mmc/lingma-proxy-compose
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Release v1.4.3

Browse Source
This commit is contained in:
lutc5
2026-04-30 18:20:04 +08:00
parent a2f777a1a8
commit a02fd51c19
24 changed files with 1909 additions and 1176 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
CHANGELOG.md
View File

@@ -1,5 +1,22 @@
# Changelog
## Unreleased
- Nothing yet.
## v1.4.3 - 2026-04-30
- Added remote API timeout fallback with a configurable model order. The default order is Kimi-K2.6, MiniMax-M2.7, Qwen3-Coder, Qwen3.6-Plus, Qwen3-Max, and Qwen3-Thinking.
- Fallback only runs before any streaming bytes are sent and only uses models returned by the active `/v1/models` response.
- Changed the default request timeout from 120 seconds to 300 seconds.
- Added a desktop Settings switch and fallback model list editor.
- Added persistent desktop app state for request history, app logs, and cumulative token usage.
- Added a Dashboard token usage card and model-list specification chips for context window and capability summaries.
- Added model display to the desktop request stream table and model-aware request search.
- Fixed Dashboard "recent model" tracking so health/model-list requests no longer override the last real chat model.
- Updated architecture documentation to cover the IPC and Remote API dual-backend design.
- Disabled desktop Inspector and default context menu in production builds; local development can opt in with `LINGMA_DESKTOP_DEBUG=1`.
## v1.4.2 - 2026-04-30
- Default backend changed to remote API mode for new CLI and desktop configurations.

17
README.md
View File

@@ -13,7 +13,7 @@ The proxy now supports two backend modes:
## Current Version
The current desktop line is `v1.4.2`.
The current desktop line is `v1.4.3`.
See [CHANGELOG.md](./CHANGELOG.md) for release history.
@@ -326,6 +326,10 @@ The proxy only reports models actually exposed by your Lingma plugin. The table
Default model when the client omits `model`: `kmodel` (`Kimi-K2.6` in the remote model list).
Remote mode enables timeout fallback by default. On timeout, upstream 5xx/429, or network interruption, the proxy only switches models if no streaming bytes have been sent to the client yet. Fallback candidates are filtered against the actual `/v1/models` response, so unavailable models are skipped. Default order:
`Kimi-K2.6 -> MiniMax-M2.7 -> Qwen3-Coder -> Qwen3.6-Plus -> Qwen3-Max -> Qwen3-Thinking`
## Configuration
Default config file:
@@ -348,7 +352,16 @@ Example:
"mode": "agent",
"shell_type": "zsh",
"session_mode": "auto",
"timeout": 120,
"timeout": 300,
"remote_fallback_enabled": true,
"remote_fallback_models": [
"kmodel",
"mmodel",
"dashscope_qwen3_coder",
"dashscope_qmodel",
"dashscope_qwen_max_latest",
"dashscope_qwen_plus_20250428_thinking"
],
"cwd": "/Users/you/project",
"current_file_path": ""
}

17
README.zh-CN.md
View File

@@ -16,7 +16,7 @@
## 当前版本
当前桌面端版本线：`v1.4.2`
当前桌面端版本线：`v1.4.3`
版本更新记录见 [CHANGELOG.md](./CHANGELOG.md)。
@@ -408,6 +408,10 @@ export ANTHROPIC_API_KEY="any"
当客户端请求没有携带 `model` 字段时，代理默认使用：`kmodel`（远端模型列表里的 Kimi-K2.6）。
远端模式默认开启超时兜底。遇到请求超时、上游 5xx/429 或网络中断时，代理只会在尚未向客户端输出任何流式内容的情况下切换模型。兜底候选会先和实际 `/v1/models` 返回结果求交集，不存在或当前账号不可用的模型会自动跳过。默认顺序：
`Kimi-K2.6 -> MiniMax-M2.7 -> Qwen3-Coder -> Qwen3.6-Plus -> Qwen3-Max -> Qwen3-Thinking`
## 配置文件
默认读取：
@@ -430,7 +434,16 @@ export ANTHROPIC_API_KEY="any"
"mode": "agent",
"shell_type": "zsh",
"session_mode": "auto",
"timeout": 120,
"timeout": 300,
"remote_fallback_enabled": true,
"remote_fallback_models": [
"kmodel",
"mmodel",
"dashscope_qwen3_coder",
"dashscope_qmodel",
"dashscope_qwen_max_latest",
"dashscope_qwen_plus_20250428_thinking"
],
"cwd": "/Users/tiancheng/project",
"current_file_path": ""
}

52
cmd/lingma-ipc-proxy/main.go
View File

@@ -38,6 +38,8 @@ type fileConfig struct {
ShellType string `json:"shell_type"`
SessionMode string `json:"session_mode"`
TimeoutSeconds int `json:"timeout"`
RemoteFallbackEnabled *bool `json:"remote_fallback_enabled"`
RemoteFallbackModels []string `json:"remote_fallback_models"`
}
func main() {
@@ -98,7 +100,9 @@ func loadConfig() (service.Config, string) {
Model: "kmodel",
ShellType: defaultShellType(),
SessionMode: service.SessionModeAuto,
Timeout: 120 * time.Second,
Timeout: 300 * time.Second,
RemoteFallbackEnabled: true,
RemoteFallbackModels: service.DefaultRemoteFallbackModels(),
}
configPath, configLoaded := resolveConfigPath()
@@ -127,6 +131,8 @@ func loadConfig() (service.Config, string) {
model := flag.String("model", cfg.Model, "Default Lingma model when API request omits model")
shellType := flag.String("shell-type", cfg.ShellType, "Shell type sent through ACP meta")
timeoutSeconds := flag.Int("timeout", int(cfg.Timeout/time.Second), "Per-request timeout in seconds")
remoteFallbackEnabled := flag.Bool("remote-fallback", cfg.RemoteFallbackEnabled, "Enable remote timeout/5xx fallback to the next available model")
remoteFallbackModels := flag.String("remote-fallback-models", strings.Join(cfg.RemoteFallbackModels, ","), "Comma-separated remote fallback model IDs")
sessionMode := flag.String("session-mode", string(cfg.SessionMode), "Session mode: auto, fresh, reuse")
config := flag.String("config", valueOr(configPath, filepath.Join(currentDir(), "lingma-ipc-proxy.json")), "Path to JSON config file")
flag.Parse()
@@ -151,6 +157,8 @@ func loadConfig() (service.Config, string) {
cfg.ShellType = strings.TrimSpace(*shellType)
cfg.SessionMode = parsedSessionMode
cfg.Timeout = time.Duration(*timeoutSeconds) * time.Second
cfg.RemoteFallbackEnabled = *remoteFallbackEnabled
cfg.RemoteFallbackModels = splitCSV(*remoteFallbackModels)
if configLoaded {
configPath = finalConfigPath
@@ -236,6 +244,12 @@ func overlayFileConfig(dst *service.Config, src fileConfig) {
if src.TimeoutSeconds > 0 {
dst.Timeout = time.Duration(src.TimeoutSeconds) * time.Second
}
if src.RemoteFallbackEnabled != nil {
dst.RemoteFallbackEnabled = *src.RemoteFallbackEnabled
}
if len(src.RemoteFallbackModels) > 0 {
dst.RemoteFallbackModels = cleanStringSlice(src.RemoteFallbackModels)
}
}
func overlayEnvConfig(dst *service.Config) {
@@ -287,6 +301,12 @@ func overlayEnvConfig(dst *service.Config) {
if value := envInt("LINGMA_PROXY_TIMEOUT_SECONDS", 0); value > 0 {
dst.Timeout = time.Duration(value) * time.Second
}
if value, ok := envBool("LINGMA_REMOTE_FALLBACK_ENABLED"); ok {
dst.RemoteFallbackEnabled = value
}
if value := strings.TrimSpace(os.Getenv("LINGMA_REMOTE_FALLBACK_MODELS")); value != "" {
dst.RemoteFallbackModels = splitCSV(value)
}
}
func parseSessionMode(value string) service.SessionMode {
@@ -349,6 +369,36 @@ func envInt(key string, fallback int) int {
return fallback
}
func envBool(key string) (bool, bool) {
value := strings.ToLower(strings.TrimSpace(os.Getenv(key)))
switch value {
case "1", "true", "yes", "on":
return true, true
case "0", "false", "no", "off":
return false, true
default:
return false, false
}
}
func splitCSV(value string) []string {
return cleanStringSlice(strings.Split(value, ","))
}
func cleanStringSlice(values []string) []string {
out := make([]string, 0, len(values))
seen := map[string]bool{}
for _, value := range values {
item := strings.TrimSpace(value)
if item == "" || seen[item] {
continue
}
seen[item] = true
out = append(out, item)
}
return out
}
func currentDir() string {
if wd, err := os.Getwd(); err == nil {
return wd

11
config.example.json
View File

@@ -6,7 +6,16 @@
"mode": "chat",
"model": "kmodel",
"session_mode": "auto",
"timeout": 120,
"timeout": 300,
"remote_fallback_enabled": true,
"remote_fallback_models": [
"kmodel",
"mmodel",
"dashscope_qwen3_coder",
"dashscope_qmodel",
"dashscope_qwen_max_latest",
"dashscope_qwen_plus_20250428_thinking"
],
"cwd": "C:/Workspace/Personal/lingma-ipc-proxy",
"shell_type": "powershell",
"current_file_path": "",

337
desktop/app.go
View File

@@ -32,10 +32,30 @@ type RequestRecord struct {
StatusCode int `json:"statusCode"`
Duration string `json:"duration"`
Size string `json:"size,omitempty"`
InputTokens int `json:"inputTokens,omitempty"`
OutputTokens int `json:"outputTokens,omitempty"`
TotalTokens int `json:"totalTokens,omitempty"`
ReqBody string `json:"reqBody,omitempty"`
RespBody string `json:"respBody,omitempty"`
}
type AppLog struct {
Time string `json:"time"`
Level string `json:"level"`
Message string `json:"message"`
}
type TokenStats struct {
TotalRequests int `json:"totalRequests"`
SuccessRequests int `json:"successRequests"`
InputTokens int `json:"inputTokens"`
OutputTokens int `json:"outputTokens"`
TotalTokens int `json:"totalTokens"`
ByModel map[string]int `json:"byModel,omitempty"`
LastModel string `json:"lastModel,omitempty"`
LastUpdated string `json:"lastUpdated,omitempty"`
}
type App struct {
ctx context.Context
@@ -49,6 +69,8 @@ type App struct {
quitHint time.Time
models []ModelInfo
requests []RequestRecord
logs []AppLog
stats TokenStats
}
// ModelInfo represents a model returned by /v1/models
@@ -96,6 +118,9 @@ func NewApp() *App {
func (a *App) startup(ctx context.Context) {
a.ctx = ctx
a.cfg = defaultConfig()
if err := a.loadAppState(); err != nil {
runtime.LogWarningf(a.ctx, "failed to load app state: %v", err)
}
// Auto-save default config on first run so users can find/edit it later
if err := a.saveConfig(a.cfg); err != nil {
@@ -208,10 +233,19 @@ func (a *App) forceQuit() {
}
func (a *App) emitLog(level string, message string) {
runtime.EventsEmit(a.ctx, "log", map[string]string{
"level": level,
"message": message,
})
entry := AppLog{
Time: time.Now().Format("15:04:05"),
Level: level,
Message: message,
}
a.mu.Lock()
a.logs = append(a.logs, entry)
if len(a.logs) > 2000 {
a.logs = a.logs[len(a.logs)-2000:]
}
a.saveAppStateLocked()
a.mu.Unlock()
runtime.EventsEmit(a.ctx, "log", entry)
}
// GetStatus returns the current proxy status
@@ -347,6 +381,8 @@ func (a *App) saveConfig(cfg service.Config) error {
"shell_type": cfg.ShellType,
"session_mode": string(cfg.SessionMode),
"timeout": timeoutSec,
"remote_fallback_enabled": cfg.RemoteFallbackEnabled,
"remote_fallback_models": cfg.RemoteFallbackModels,
}
data, err := json.MarshalIndent(fileCfg, "", " ")
@@ -361,14 +397,16 @@ func (a *App) saveConfig(cfg service.Config) error {
// StartProxy starts the lingma-ipc-proxy HTTP server
func (a *App) StartProxy() error {
a.mu.Lock()
defer a.mu.Unlock()
if a.running {
a.mu.Unlock()
return fmt.Errorf("proxy already running")
}
addr := fmt.Sprintf("%s:%d", a.cfg.Host, a.cfg.Port)
svc := service.New(a.cfg)
cfg := a.cfg
a.mu.Unlock()
svc := service.New(cfg)
warmupCtx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
if err := svc.Warmup(warmupCtx); err != nil {
@@ -382,23 +420,32 @@ func (a *App) StartProxy() error {
server := httpapi.NewServer(addr, svc)
server.OnRequest = func(method, path string, statusCode int, duration time.Duration, reqBody, respBody string) {
a.mu.Lock()
a.requests = append(a.requests, RequestRecord{
inputTokens, outputTokens := extractTokenUsage(respBody)
model := extractRequestModel(reqBody)
record := RequestRecord{
Time: time.Now().Format("15:04:05"),
Method: method,
Path: path,
Model: extractRequestModel(reqBody),
Model: model,
StatusCode: statusCode,
Duration: duration.Round(time.Millisecond).String(),
Size: formatPayloadSize(len(reqBody) + len(respBody)),
InputTokens: inputTokens,
OutputTokens: outputTokens,
TotalTokens: inputTokens + outputTokens,
ReqBody: reqBody,
RespBody: respBody,
})
if len(a.requests) > 100 {
a.requests = a.requests[len(a.requests)-100:]
}
a.mu.Lock()
a.requests = append(a.requests, record)
if len(a.requests) > 2000 {
a.requests = a.requests[len(a.requests)-2000:]
}
a.accumulateTokenStatsLocked(record)
a.saveAppStateLocked()
a.mu.Unlock()
runtime.EventsEmit(a.ctx, "requests:updated", a.GetRequests())
runtime.EventsEmit(a.ctx, "usage:updated", a.GetTokenStats())
}
// Check if the port is available before claiming we're running
@@ -420,10 +467,16 @@ func (a *App) StartProxy() error {
}
}()
a.mu.Lock()
if a.running {
a.mu.Unlock()
return fmt.Errorf("proxy already running")
}
a.server = server
a.addr = addr
a.running = true
a.startedAt = time.Now()
a.mu.Unlock()
msg := fmt.Sprintf("Proxy started on http://%s", addr)
runtime.LogInfof(a.ctx, msg)
@@ -435,8 +488,24 @@ func (a *App) StartProxy() error {
return nil
}
// ClearLogs is a no-op backend helper (logs are kept in frontend memory)
func (a *App) ClearLogs() {}
func (a *App) GetLogs() []AppLog {
a.mu.RLock()
defer a.mu.RUnlock()
out := make([]AppLog, len(a.logs))
copy(out, a.logs)
for i, j := 0, len(out)-1; i < j; i, j = i+1, j-1 {
out[i], out[j] = out[j], out[i]
}
return out
}
func (a *App) ClearLogs() {
a.mu.Lock()
a.logs = nil
a.saveAppStateLocked()
a.mu.Unlock()
runtime.EventsEmit(a.ctx, "logs:updated", a.GetLogs())
}
// StopProxy stops the proxy server
func (a *App) StopProxy() error {
@@ -493,10 +562,21 @@ func (a *App) GetRequests() []RequestRecord {
func (a *App) ClearRequests() {
a.mu.Lock()
a.requests = nil
a.saveAppStateLocked()
a.mu.Unlock()
a.emitLog("info", "Request history cleared")
}
func (a *App) GetTokenStats() TokenStats {
a.mu.RLock()
defer a.mu.RUnlock()
stats := a.stats
if stats.ByModel != nil {
stats.ByModel = cloneIntMap(stats.ByModel)
}
return stats
}
// RefreshModels probes the running proxy for the latest model list.
func (a *App) RefreshModels() ([]ModelInfo, error) {
a.mu.RLock()
@@ -614,6 +694,207 @@ func formatPayloadSize(bytes int) string {
return fmt.Sprintf("%d B", bytes)
}
type appStateFile struct {
Requests []RequestRecord `json:"requests"`
Logs []AppLog `json:"logs"`
Stats TokenStats `json:"stats"`
}
func (a *App) loadAppState() error {
path, err := appStatePath()
if err != nil {
return err
}
data, err := os.ReadFile(path)
if err != nil {
if os.IsNotExist(err) {
return nil
}
return err
}
var state appStateFile
if err := json.Unmarshal(data, &state); err != nil {
return err
}
a.mu.Lock()
defer a.mu.Unlock()
a.requests = state.Requests
a.logs = state.Logs
a.stats = state.Stats
if a.stats.ByModel == nil {
a.stats.ByModel = map[string]int{}
}
a.reconcileTokenStatsLocked()
return nil
}
func (a *App) saveAppStateLocked() {
path, err := appStatePath()
if err != nil {
runtime.LogWarningf(a.ctx, "resolve app state path failed: %v", err)
return
}
if err := os.MkdirAll(filepath.Dir(path), 0755); err != nil {
runtime.LogWarningf(a.ctx, "create app state dir failed: %v", err)
return
}
state := appStateFile{
Requests: a.requests,
Logs: a.logs,
Stats: a.stats,
}
data, err := json.MarshalIndent(state, "", " ")
if err != nil {
runtime.LogWarningf(a.ctx, "marshal app state failed: %v", err)
return
}
if err := os.WriteFile(path, data, 0644); err != nil {
runtime.LogWarningf(a.ctx, "write app state failed: %v", err)
}
}
func appStatePath() (string, error) {
home, err := os.UserHomeDir()
if err != nil {
return "", err
}
return filepath.Join(home, ".config", "lingma-ipc-proxy", "app-state.json"), nil
}
func (a *App) accumulateTokenStatsLocked(record RequestRecord) {
a.stats.TotalRequests++
if record.StatusCode >= 200 && record.StatusCode < 300 {
a.stats.SuccessRequests++
}
a.stats.InputTokens += record.InputTokens
a.stats.OutputTokens += record.OutputTokens
a.stats.TotalTokens += record.TotalTokens
if a.stats.ByModel == nil {
a.stats.ByModel = map[string]int{}
}
model := strings.TrimSpace(record.Model)
if model == "" {
model = "-"
}
if record.TotalTokens > 0 {
a.stats.ByModel[model] += record.TotalTokens
if isUsageBearingRequest(record.Path) && model != "-" {
a.stats.LastModel = model
}
}
a.stats.LastUpdated = time.Now().Format(time.RFC3339)
}
func (a *App) reconcileTokenStatsLocked() {
if a.stats.ByModel == nil {
a.stats.ByModel = map[string]int{}
}
a.stats.LastModel = ""
for i := len(a.requests) - 1; i >= 0; i-- {
record := a.requests[i]
model := strings.TrimSpace(record.Model)
if model == "" || record.TotalTokens <= 0 || !isUsageBearingRequest(record.Path) {
continue
}
a.stats.LastModel = model
break
}
}
func isUsageBearingRequest(path string) bool {
switch strings.TrimSpace(path) {
case "/v1/messages", "/v1/chat/completions", "/v1/completions":
return true
default:
return false
}
}
func cloneIntMap(src map[string]int) map[string]int {
out := make(map[string]int, len(src))
for k, v := range src {
out[k] = v
}
return out
}
func extractTokenUsage(respBody string) (int, int) {
if strings.TrimSpace(respBody) == "" {
return 0, 0
}
input, output := extractUsageFromJSON(respBody)
if input != 0 || output != 0 {
return input, output
}
for _, line := range strings.Split(respBody, "\n") {
line = strings.TrimSpace(line)
if !strings.HasPrefix(line, "data:") {
continue
}
payload := strings.TrimSpace(strings.TrimPrefix(line, "data:"))
if payload == "" || payload == "[DONE]" {
continue
}
in, out := extractUsageFromJSON(payload)
if in > 0 {
input = in
}
if out > 0 {
output = out
}
}
return input, output
}
func extractUsageFromJSON(raw string) (int, int) {
var payload any
if err := json.Unmarshal([]byte(raw), &payload); err != nil {
return 0, 0
}
usage, ok := findUsageMap(payload)
if !ok {
return 0, 0
}
input := intFromAny(usage["input_tokens"]) + intFromAny(usage["prompt_tokens"])
output := intFromAny(usage["output_tokens"]) + intFromAny(usage["completion_tokens"])
return input, output
}
func findUsageMap(value any) (map[string]any, bool) {
switch typed := value.(type) {
case map[string]any:
if usage, ok := typed["usage"].(map[string]any); ok {
return usage, true
}
for _, child := range typed {
if usage, ok := findUsageMap(child); ok {
return usage, true
}
}
case []any:
for _, child := range typed {
if usage, ok := findUsageMap(child); ok {
return usage, true
}
}
}
return nil, false
}
func intFromAny(value any) int {
switch typed := value.(type) {
case float64:
return int(typed)
case int:
return typed
case json.Number:
n, _ := typed.Int64()
return int(n)
default:
return 0
}
}
func defaultConfig() service.Config {
cfg := service.Config{
Host: "127.0.0.1",
@@ -625,7 +906,9 @@ func defaultConfig() service.Config {
Model: "kmodel",
ShellType: defaultShellType(),
SessionMode: service.SessionModeAuto,
Timeout: 120 * time.Second,
Timeout: 300 * time.Second,
RemoteFallbackEnabled: true,
RemoteFallbackModels: service.DefaultRemoteFallbackModels(),
}
// Try to load config file from multiple locations
@@ -650,6 +933,8 @@ func defaultConfig() service.Config {
ShellType string `json:"shell_type"`
SessionMode string `json:"session_mode"`
TimeoutSeconds int `json:"timeout"`
RemoteFallbackEnabled *bool `json:"remote_fallback_enabled"`
RemoteFallbackModels []string `json:"remote_fallback_models"`
}
if err := json.Unmarshal(data, &fileCfg); err == nil {
if fileCfg.Host != "" {
@@ -702,6 +987,12 @@ func defaultConfig() service.Config {
if fileCfg.TimeoutSeconds > 0 {
cfg.Timeout = time.Duration(fileCfg.TimeoutSeconds) * time.Second
}
if fileCfg.RemoteFallbackEnabled != nil {
cfg.RemoteFallbackEnabled = *fileCfg.RemoteFallbackEnabled
}
if len(fileCfg.RemoteFallbackModels) > 0 {
cfg.RemoteFallbackModels = cleanConfigStrings(fileCfg.RemoteFallbackModels)
}
}
break // loaded successfully
}
@@ -732,6 +1023,20 @@ func maskIdentifier(value string) string {
return string(runes[:4]) + "..." + string(runes[len(runes)-4:])
}
func cleanConfigStrings(values []string) []string {
out := make([]string, 0, len(values))
seen := map[string]bool{}
for _, value := range values {
item := strings.TrimSpace(value)
if item == "" || seen[item] {
continue
}
seen[item] = true
out = append(out, item)
}
return out
}
func configSearchPaths() []string {
var paths []string
// 1. Executable directory (for dev / portable mode)

23
desktop/frontend/src/App.vue
View File

@@ -6,7 +6,7 @@ import Models from './views/Models.vue'
import Requests from './views/Requests.vue'
import Settings from './views/Settings.vue'
import { EventsOff, EventsOn } from '../wailsjs/runtime'
import { GetStatus, HideWindow, MinimizeWindow } from '../wailsjs/go/main/App.js'
import { ClearLogs, GetLogs, GetStatus, HideWindow, MinimizeWindow } from '../wailsjs/go/main/App.js'
import lingmaIcon from './assets/images/lingma-icon.png'
const currentTab = ref('dashboard')
@@ -42,8 +42,13 @@ function showToast(message) {
}, 2200)
}
function clearLocalLogs() {
async function clearLocalLogs() {
try {
await ClearLogs()
logs.value = []
} catch (e) {
logs.value = []
}
}
function setStatus(nextStatus) {
@@ -158,14 +163,25 @@ onMounted(() => {
systemThemeQuery?.addEventListener?.('change', applyTheme)
applyTheme()
refreshStatus()
GetLogs().then((items) => {
logs.value = Array.isArray(items) ? items : []
}).catch(() => {})
safeEventsOn('models:updated', (data) => {
status.value.models = Array.isArray(data) ? data.length : status.value.models
addLog('info', `模型列表已更新：${status.value.models} 个模型`)
})
safeEventsOn('log', (data) => {
if (data.time && data.message !== undefined) {
logs.value.unshift(data)
if (logs.value.length > 500) logs.value = logs.value.slice(0, 500)
} else {
addLog(data.level || 'info', data.message || '')
}
refreshStatus()
})
safeEventsOn('logs:updated', (data) => {
logs.value = Array.isArray(data) ? data : []
})
safeEventsOn('quit:confirm', (message) => {
showToast(message || '再按一次退出快捷键将停止代理并退出应用')
})
@@ -183,6 +199,7 @@ onUnmounted(() => {
systemThemeQuery?.removeEventListener?.('change', applyTheme)
safeEventsOff('models:updated')
safeEventsOff('log')
safeEventsOff('logs:updated')
safeEventsOff('quit:confirm')
safeEventsOff('status:updated')
safeEventsOff('requests:updated')
@@ -222,7 +239,7 @@ onUnmounted(() => {
<span class="status-dot" :class="{ running: status.running }"></span>
<div>
<strong>{{ status.running ? 'Proxy Running' : 'Proxy Stopped' }}</strong>
<small>v1.4.2</small>
<small>v1.4.3</small>
</div>
</div>
</aside>

71
desktop/frontend/src/components/HelloWorld.vue
View File

@@ -1,71 +0,0 @@
<script lang="ts" setup>
import {reactive} from 'vue'
import {Greet} from '../../wailsjs/go/main/App'
const data = reactive({
name: "",
resultText: "Please enter your name below 👇",
})
function greet() {
Greet(data.name).then(result => {
data.resultText = result
})
}
</script>
<template>
<main>
<div id="result" class="result">{{ data.resultText }}</div>
<div id="input" class="input-box">
<input id="name" v-model="data.name" autocomplete="off" class="input" type="text"/>
<button class="btn" @click="greet">Greet</button>
</div>
</main>
</template>
<style scoped>
.result {
height: 20px;
line-height: 20px;
margin: 1.5rem auto;
}
.input-box .btn {
width: 60px;
height: 30px;
line-height: 30px;
border-radius: 3px;
border: none;
margin: 0 0 0 20px;
padding: 0 8px;
cursor: pointer;
}
.input-box .btn:hover {
background-image: linear-gradient(to top, #cfd9df 0%, #e2ebf0 100%);
color: #333333;
}
.input-box .input {
border: none;
border-radius: 3px;
outline: none;
height: 30px;
line-height: 30px;
padding: 0 10px;
background-color: rgba(240, 240, 240, 1);
-webkit-font-smoothing: antialiased;
}
.input-box .input:hover {
border: none;
background-color: rgba(255, 255, 255, 1);
}
.input-box .input:focus {
border: none;
background-color: rgba(255, 255, 255, 1);
}
</style>

495
desktop/frontend/src/style.css
View File

@@ -1,5 +1,12 @@
:root {
font-family: Inter, ui-sans-serif, -apple-system, BlinkMacSystemFont, "SF Pro Text", "Segoe UI", sans-serif;
font-family:
Inter,
ui-sans-serif,
-apple-system,
BlinkMacSystemFont,
'SF Pro Text',
'Segoe UI',
sans-serif;
color: #172033;
background: #eef2f6;
font-synthesis: none;
@@ -26,7 +33,7 @@
--radius: 8px;
}
:root[data-theme="dark"] {
:root[data-theme='dark'] {
color: #edf3ff;
background: #111827;
--bg: #111827;
@@ -68,7 +75,7 @@ body {
background: var(--bg);
}
:root[data-theme="dark"] body {
:root[data-theme='dark'] body {
background: var(--bg);
}
@@ -106,7 +113,7 @@ button {
box-shadow: none;
}
:root[data-theme="dark"] .app-shell {
:root[data-theme='dark'] .app-shell {
border-color: rgba(148, 163, 184, 0.22);
background: rgba(16, 24, 36, 0.78);
}
@@ -123,7 +130,7 @@ button {
box-shadow: inset -1px 0 0 rgba(125, 139, 158, 0.16);
}
:root[data-theme="dark"] .sidebar {
:root[data-theme='dark'] .sidebar {
border-right-color: rgba(148, 163, 184, 0.14);
background: linear-gradient(180deg, rgba(28, 39, 56, 0.7), rgba(18, 27, 40, 0.66));
box-shadow: inset -1px 0 0 rgba(148, 163, 184, 0.12);
@@ -145,13 +152,13 @@ button {
background: rgba(255, 255, 255, 0.58);
}
:root[data-theme="dark"] .brand:hover,
:root[data-theme="dark"] .nav-item:hover,
:root[data-theme="dark"] .sidebar-status {
:root[data-theme='dark'] .brand:hover,
:root[data-theme='dark'] .nav-item:hover,
:root[data-theme='dark'] .sidebar-status {
background: rgba(255, 255, 255, 0.08);
}
:root[data-theme="dark"] .nav-item {
:root[data-theme='dark'] .nav-item {
color: #aebbd0;
}
@@ -224,7 +231,7 @@ button {
box-shadow: inset 0 0 0 1px rgba(37, 99, 235, 0.12);
}
:root[data-theme="dark"] .nav-item.active {
:root[data-theme='dark'] .nav-item.active {
color: #d8e6ff;
background: rgba(67, 111, 190, 0.24);
box-shadow: inset 0 0 0 1px rgba(105, 161, 255, 0.18);
@@ -285,13 +292,13 @@ button {
min-height: 46px;
padding: 0 16px;
border-bottom: 1px solid rgba(112, 128, 148, 0.18);
background: rgba(255, 255, 255, 0.58);
backdrop-filter: blur(20px) saturate(1.08);
background: #f6f9fd;
backdrop-filter: none;
}
:root[data-theme="dark"] .topbar {
:root[data-theme='dark'] .topbar {
border-bottom-color: rgba(148, 163, 184, 0.14);
background: rgba(20, 30, 45, 0.66);
background: #162131;
}
.topbar-spacer {
@@ -372,10 +379,10 @@ button {
backdrop-filter: blur(18px) saturate(1.12);
}
:root[data-theme="dark"] .glass-panel,
:root[data-theme="dark"] .metric,
:root[data-theme="dark"] .table-panel,
:root[data-theme="dark"] .config-panel {
:root[data-theme='dark'] .glass-panel,
:root[data-theme='dark'] .metric,
:root[data-theme='dark'] .table-panel,
:root[data-theme='dark'] .config-panel {
border-color: rgba(148, 163, 184, 0.14);
background: var(--surface);
}
@@ -583,6 +590,10 @@ button {
gap: 18px;
}
.settings-grid {
align-items: start;
}
.grid-3 {
display: grid;
grid-template-columns: repeat(3, minmax(0, 1fr));
@@ -591,10 +602,11 @@ button {
.dashboard-grid {
display: grid;
align-items: stretch;
grid-template-columns: minmax(0, 1fr) minmax(0, 0.95fr) minmax(300px, 0.95fr);
grid-template-areas:
"health models config"
"requests requests config";
'health models config'
'requests requests usage';
gap: 12px;
}
@@ -604,16 +616,109 @@ button {
.area-models {
grid-area: models;
min-height: 0;
}
.area-config {
grid-area: config;
}
.compact-header {
margin-bottom: 10px;
}
.compact-header p {
margin-top: 4px;
}
.area-usage {
grid-area: usage;
}
.area-requests {
grid-area: requests;
}
.usage-grid {
display: grid;
grid-template-columns: repeat(2, minmax(0, 1fr));
gap: 10px;
}
.usage-grid div {
min-width: 0;
padding: 12px;
border: 1px solid var(--line);
border-radius: 8px;
background: var(--surface-soft);
}
.usage-grid label {
display: block;
margin-bottom: 6px;
color: var(--muted);
font-size: 11px;
font-weight: 680;
}
.usage-grid strong {
display: block;
overflow: hidden;
color: var(--text);
font-size: 20px;
line-height: 1.15;
text-overflow: ellipsis;
}
.usage-foot {
display: flex;
flex-wrap: wrap;
gap: 8px 14px;
margin-top: 12px;
color: var(--muted);
font-size: 12px;
}
.config-summary {
display: grid;
grid-template-columns: repeat(2, minmax(0, 1fr));
gap: 10px;
}
.config-summary-item {
min-width: 0;
padding: 12px;
border: 1px solid var(--line);
border-radius: 8px;
background: var(--surface-soft);
}
.config-summary-item label {
display: block;
margin-bottom: 6px;
color: var(--muted);
font-size: 11px;
font-weight: 680;
}
.config-summary-item strong {
display: block;
overflow: hidden;
color: var(--text);
font-size: 13px;
line-height: 1.3;
text-overflow: ellipsis;
white-space: nowrap;
}
.config-summary-item.span-2 {
grid-column: 1 / -1;
}
.compact-link {
margin-top: 10px;
}
.activity-chart {
display: grid;
grid-template-columns: repeat(36, minmax(3px, 1fr));
@@ -637,13 +742,13 @@ button {
white-space: nowrap;
}
:root[data-theme="dark"] .activity-chart,
:root[data-theme="dark"] .data-table th,
:root[data-theme="dark"] .field input,
:root[data-theme="dark"] .field textarea,
:root[data-theme="dark"] .search-input,
:root[data-theme="dark"] .detail-panel pre,
:root[data-theme="dark"] .code-block {
:root[data-theme='dark'] .activity-chart,
:root[data-theme='dark'] .data-table th,
:root[data-theme='dark'] .field input,
:root[data-theme='dark'] .field textarea,
:root[data-theme='dark'] .search-input,
:root[data-theme='dark'] .detail-panel pre,
:root[data-theme='dark'] .code-block {
color: var(--text);
border-color: var(--line);
background: rgba(15, 23, 42, 0.74);
@@ -719,8 +824,8 @@ button {
box-shadow: inset 0 0 0 1px rgba(37, 99, 235, 0.12);
}
:root[data-theme="dark"] .model-choice:hover,
:root[data-theme="dark"] .model-choice:focus-visible {
:root[data-theme='dark'] .model-choice:hover,
:root[data-theme='dark'] .model-choice:focus-visible {
color: #f3f7ff;
border-color: rgba(105, 161, 255, 0.38);
background: rgba(72, 118, 214, 0.34);
@@ -728,7 +833,48 @@ button {
.models-list .model-row,
.model-list-row {
grid-template-columns: 22px minmax(220px, 1fr) auto;
grid-template-columns: 22px minmax(220px, 1fr) minmax(260px, auto);
}
.model-specs {
display: flex;
flex-wrap: wrap;
justify-content: flex-end;
gap: 6px;
}
.spec-chip {
display: inline-flex;
min-height: 22px;
align-items: center;
padding: 0 8px;
border: 1px solid var(--line);
border-radius: 7px;
color: var(--muted);
background: var(--surface-soft);
font-size: 11px;
font-weight: 680;
white-space: nowrap;
}
.spec-chip.strong {
color: #0d6a41;
border-color: rgba(24, 160, 88, 0.18);
background: var(--green-soft);
}
.spec-chip.muted-chip {
color: #8a5a08;
border-color: rgba(217, 119, 6, 0.16);
background: var(--warn-soft);
}
:root[data-theme='dark'] .spec-chip.strong {
color: #7ee0aa;
}
:root[data-theme='dark'] .spec-chip.muted-chip {
color: #ffd27a;
}
.model-brand-icon {
@@ -744,6 +890,8 @@ button {
display: flex;
min-height: 0;
flex-direction: column;
overflow: hidden;
height: 295px;
}
.model-card-list,
@@ -754,7 +902,14 @@ button {
}
.model-card-list {
max-height: 248px;
flex: 1 1 auto;
max-height: none;
scrollbar-width: none;
}
.model-card-list::-webkit-scrollbar {
width: 0;
height: 0;
}
.model-page-list {
@@ -870,6 +1025,66 @@ button {
border-bottom: 1px solid var(--line);
}
.toolbar-header {
margin: 0;
display: flex;
align-items: baseline;
gap: 8px;
}
.toolbar-count {
font-size: 12px;
font-weight: normal;
white-space: nowrap;
}
.toolbar-search-wrap {
display: flex;
align-items: center;
gap: 12px;
flex: 1;
}
.toolbar-search-input {
max-width: 300px;
width: 100%;
}
.btn-sm-outline {
padding: 4px 10px;
font-size: 12px;
background: transparent;
border: 1px solid var(--line);
border-radius: 6px;
cursor: pointer;
color: var(--text);
display: inline-flex;
align-items: center;
justify-content: center;
transition: all 0.16s ease;
}
.btn-sm-outline:hover {
background: rgba(0, 0, 0, 0.05);
}
.btn-sm-outline:disabled {
opacity: 0.6;
cursor: not-allowed;
}
.btn-sm-outline i {
margin-left: 2px;
}
:root[data-theme='dark'] .btn-sm-outline {
color: #dce8fb;
}
:root[data-theme='dark'] .btn-sm-outline:hover {
background: rgba(255, 255, 255, 0.1);
}
.table-scroll {
flex: 0 0 auto;
max-height: none;
@@ -893,7 +1108,7 @@ button {
.area-requests .table-scroll {
min-height: 0;
max-height: 260px;
max-height: 211px;
overflow: auto;
}
@@ -932,10 +1147,12 @@ button {
}
.data-table tbody tr {
height: var(--request-row-height, 64px);
height: var(--request-row-height, 42px);
cursor: pointer;
background: rgba(255, 255, 255, 0.34);
transition: background-color 140ms ease, box-shadow 140ms ease;
transition:
background-color 140ms ease,
box-shadow 140ms ease;
}
.data-table tbody tr:hover {
@@ -947,23 +1164,23 @@ button {
box-shadow: inset 3px 0 0 var(--blue);
}
:root[data-theme="dark"] .data-table {
:root[data-theme='dark'] .data-table {
background: rgba(15, 23, 42, 0.8);
}
:root[data-theme="dark"] .data-table th {
:root[data-theme='dark'] .data-table th {
background: rgba(15, 23, 42, 0.96);
}
:root[data-theme="dark"] .data-table tbody tr {
:root[data-theme='dark'] .data-table tbody tr {
background: rgba(20, 31, 48, 0.7);
}
:root[data-theme="dark"] .data-table tbody tr:hover {
:root[data-theme='dark'] .data-table tbody tr:hover {
background: rgba(45, 65, 96, 0.9);
}
:root[data-theme="dark"] .data-table tbody tr.selected {
:root[data-theme='dark'] .data-table tbody tr.selected {
background: rgba(38, 65, 112, 0.96);
box-shadow: inset 3px 0 0 #67a1ff;
}
@@ -1001,8 +1218,7 @@ button {
background: var(--red-soft);
}
.link-row,
.table-footer button {
.link-row {
display: flex;
align-items: center;
justify-content: space-between;
@@ -1014,24 +1230,10 @@ button {
cursor: pointer;
}
:root[data-theme="dark"] .link-row,
:root[data-theme="dark"] .table-footer button {
:root[data-theme='dark'] .link-row {
color: #dce8fb;
}
.table-footer {
display: flex;
align-items: center;
justify-content: space-between;
padding: 10px 14px;
color: var(--muted);
font-size: 12px;
}
.table-footer button {
width: auto;
gap: 8px;
}
.method-chip {
color: #334155;
@@ -1047,7 +1249,10 @@ button {
min-height: 32px;
border-radius: 8px;
cursor: pointer;
transition: transform 0.16s ease, background 0.16s ease, box-shadow 0.16s ease;
transition:
transform 0.16s ease,
background 0.16s ease,
box-shadow 0.16s ease;
}
.primary-button {
@@ -1132,12 +1337,92 @@ button:disabled {
gap: 6px;
}
.settings-fieldset {
min-width: 0;
padding: 0;
margin: 0;
border: 0;
}
.settings-fieldset:disabled {
opacity: 0.56;
}
.compact-hint {
margin-bottom: 14px;
}
.compact-form-grid {
row-gap: 14px;
}
.field label {
color: var(--muted);
font-size: 12px;
font-weight: 680;
}
.switch-field {
display: flex;
align-items: center;
justify-content: space-between;
gap: 14px;
}
.switch-field p {
margin: 4px 0 0;
color: var(--muted);
font-size: 12px;
line-height: 1.45;
}
.switch {
position: relative;
flex: 0 0 auto;
display: inline-flex;
width: 44px;
height: 26px;
}
.switch input {
position: absolute;
inset: 0;
opacity: 0;
}
.switch span {
position: absolute;
inset: 0;
border: 1px solid var(--line-strong);
border-radius: 999px;
background: rgba(148, 163, 184, 0.28);
transition:
background 0.16s ease,
border-color 0.16s ease;
}
.switch span::after {
content: '';
position: absolute;
top: 3px;
left: 3px;
width: 18px;
height: 18px;
border-radius: 999px;
background: white;
box-shadow: 0 2px 8px rgba(15, 23, 42, 0.2);
transition: transform 0.16s ease;
}
.switch input:checked + span {
border-color: rgba(37, 99, 235, 0.72);
background: #2563eb;
}
.switch input:checked + span::after {
transform: translateX(18px);
}
.field input,
.field textarea,
.search-input {
@@ -1151,6 +1436,14 @@ button:disabled {
outline: none;
}
.field .switch input {
width: auto;
min-height: 0;
padding: 0;
border: 0;
background: transparent;
}
.field textarea {
min-height: 78px;
padding-top: 9px;
@@ -1230,14 +1523,14 @@ button:disabled {
background: var(--blue-soft);
}
:root[data-theme="dark"] .custom-select > button {
:root[data-theme='dark'] .custom-select > button {
color: var(--text);
border-color: var(--line);
background: rgba(15, 23, 42, 0.74);
}
:root[data-theme="dark"] .select-menu button:hover,
:root[data-theme="dark"] .select-menu button.selected {
:root[data-theme='dark'] .select-menu button:hover,
:root[data-theme='dark'] .select-menu button.selected {
color: #dce9ff;
background: rgba(72, 118, 214, 0.32);
}
@@ -1262,7 +1555,7 @@ button:disabled {
.hint-box code {
color: var(--text);
font-family: "SF Mono", ui-monospace, Menlo, Consolas, monospace;
font-family: 'SF Mono', ui-monospace, Menlo, Consolas, monospace;
font-size: 12px;
}
@@ -1276,7 +1569,7 @@ button:disabled {
background: rgba(255, 255, 255, 0.54);
}
:root[data-theme="dark"] .detect-card {
:root[data-theme='dark'] .detect-card {
background: rgba(15, 23, 42, 0.52);
}
@@ -1326,7 +1619,7 @@ button:disabled {
margin: 0;
color: var(--text);
overflow-wrap: anywhere;
font-family: "SF Mono", ui-monospace, Menlo, Consolas, monospace;
font-family: 'SF Mono', ui-monospace, Menlo, Consolas, monospace;
font-size: 12px;
line-height: 1.45;
}
@@ -1377,7 +1670,7 @@ button:disabled {
user-select: text;
}
:root[data-theme="dark"] .detail-panel {
:root[data-theme='dark'] .detail-panel {
background: rgba(12, 18, 30, 0.96);
}
@@ -1423,7 +1716,7 @@ button:disabled {
-webkit-user-select: text;
user-select: text;
background: rgba(255, 255, 255, 0.82);
font-family: "SF Mono", ui-monospace, Menlo, Consolas, monospace;
font-family: 'SF Mono', ui-monospace, Menlo, Consolas, monospace;
font-size: 12px;
line-height: 1.55;
overflow-wrap: anywhere;
@@ -1525,28 +1818,28 @@ button:disabled {
border-color: rgba(44, 111, 231, 0.38);
}
:root[data-theme="dark"] .json-key {
:root[data-theme='dark'] .json-key {
color: #c4b5fd;
}
:root[data-theme="dark"] .json-string {
:root[data-theme='dark'] .json-string {
color: #86efac;
}
:root[data-theme="dark"] .json-number {
:root[data-theme='dark'] .json-number {
color: #93c5fd;
}
:root[data-theme="dark"] .json-boolean {
:root[data-theme='dark'] .json-boolean {
color: #fca5a5;
}
:root[data-theme="dark"] .json-null,
:root[data-theme="dark"] .json-punctuation {
:root[data-theme='dark'] .json-null,
:root[data-theme='dark'] .json-punctuation {
color: #9aa8bd;
}
:root[data-theme="dark"] .json-summary {
:root[data-theme='dark'] .json-summary {
color: #b7c3d6;
border-color: rgba(148, 163, 184, 0.24);
background: rgba(30, 41, 59, 0.78);
@@ -1563,19 +1856,17 @@ button:disabled {
height: 0;
}
:root[data-theme="dark"] .detail-panel pre,
:root[data-theme="dark"] .code-block,
:root[data-theme="dark"] .json-viewer {
:root[data-theme='dark'] .detail-panel pre,
:root[data-theme='dark'] .code-block,
:root[data-theme='dark'] .json-viewer {
color: var(--text);
border-color: var(--line);
background: rgba(17, 24, 39, 0.94);
}
.log-row {
grid-template-columns: 82px 58px minmax(0, 1fr);
font-family: "SF Mono", ui-monospace, Menlo, Consolas, monospace;
font-family: 'SF Mono', ui-monospace, Menlo, Consolas, monospace;
font-size: 12px;
-webkit-user-select: text;
user-select: text;
@@ -1636,9 +1927,10 @@ button:disabled {
.dashboard-grid {
grid-template-areas:
"health models"
"config config"
"requests requests";
'health models'
'config config'
'usage usage'
'requests requests';
}
.status-strip {
@@ -1654,34 +1946,38 @@ button:disabled {
.strip-actions {
grid-column: span 2;
}
.config-summary {
grid-template-columns: 1fr 1fr;
}
}
:root[data-theme="dark"] .strip-actions,
:root[data-theme="dark"] .secondary-button,
:root[data-theme="dark"] .ghost-button,
:root[data-theme="dark"] .icon-button,
:root[data-theme="dark"] .segmented,
:root[data-theme="dark"] .segmented button {
:root[data-theme='dark'] .strip-actions,
:root[data-theme='dark'] .secondary-button,
:root[data-theme='dark'] .ghost-button,
:root[data-theme='dark'] .icon-button,
:root[data-theme='dark'] .segmented,
:root[data-theme='dark'] .segmented button {
color: var(--text);
border-color: var(--line);
background: rgba(30, 41, 59, 0.66);
}
:root[data-theme="dark"] .strip-actions {
:root[data-theme='dark'] .strip-actions {
background: rgba(15, 23, 42, 0.78);
}
:root[data-theme="dark"] .strip-actions button {
:root[data-theme='dark'] .strip-actions button {
color: #e6eefc;
}
:root[data-theme="dark"] .strip-actions button:disabled {
:root[data-theme='dark'] .strip-actions button:disabled {
color: #a9b7cc;
background: rgba(15, 23, 42, 0.52);
opacity: 0.86;
}
:root[data-theme="dark"] .segmented button.active {
:root[data-theme='dark'] .segmented button.active {
color: #f8fbff;
background: rgba(72, 118, 214, 0.42);
}
@@ -1759,10 +2055,11 @@ button:disabled {
.dashboard-grid {
grid-template-areas:
"health"
"models"
"config"
"requests";
'health'
'models'
'config'
'usage'
'requests';
}
.status-strip {
@@ -1775,6 +2072,14 @@ button:disabled {
width: 100%;
}
.config-summary {
grid-template-columns: 1fr;
}
.config-summary-item.span-2 {
grid-column: auto;
}
.span-2 {
grid-column: auto;
}

183
desktop/frontend/src/views/Dashboard.vue
View File

@@ -1,23 +1,14 @@
<script setup>
import { computed, onMounted, onUnmounted, ref } from 'vue'
import {
GetModels,
GetConfig,
GetRequests,
GetStatus,
QuitApp,
RefreshModels,
StartProxy,
StopProxy,
} from '../../wailsjs/go/main/App.js'
import { GetModels, GetConfig, GetRequests, GetStatus, GetTokenStats, QuitApp, RefreshModels, StartProxy, StopProxy } from '../../wailsjs/go/main/App.js'
import { ClipboardSetText } from '../../wailsjs/runtime'
import { modelIcon } from '../modelIcons'
const props = defineProps({
shellStatus: {
type: Object,
default: () => ({ running: false, addr: '', models: 0 }),
},
default: () => ({ running: false, addr: '', models: 0 })
}
})
const emit = defineEmits(['log', 'status', 'notice', 'open-settings', 'open-requests', 'open-models'])
@@ -25,9 +16,11 @@ const emit = defineEmits(['log', 'status', 'notice', 'open-settings', 'open-requ
const status = ref(props.shellStatus)
const models = ref([])
const requests = ref([])
const tokenStats = ref({ totalRequests: 0, successRequests: 0, inputTokens: 0, outputTokens: 0, totalTokens: 0 })
const health = ref(null)
const config = ref({})
const loading = ref(false)
const proxyLoading = ref(false)
const modelsLoading = ref(false)
const testing = ref(false)
const now = ref(Date.now())
let interval = null
@@ -63,7 +56,7 @@ const healthStats = computed(() => {
avg,
p50: percentile(sorted, 0.5),
p95: percentile(sorted, 0.95),
max: sorted[sorted.length - 1],
max: Math.round(sorted[sorted.length - 1])
}
})
const chartBars = computed(() => {
@@ -78,17 +71,22 @@ const displayRequests = computed(() => {
})
const displayModels = computed(() => {
if (models.value.length > 0) {
return models.value.slice(0, 5).map((model) => ({ ...model, online: true }))
return models.value.map((model) => ({ ...model, online: true }))
}
return []
})
const successRate = computed(() => {
const total = Number(tokenStats.value.totalRequests || 0)
if (!total) return '0%'
return `${Math.round((Number(tokenStats.value.successRequests || 0) / total) * 100)}%`
})
function parseDurationMs(duration) {
const text = String(duration || '').trim()
if (!text) return 0
if (text.endsWith('ms')) return Number.parseFloat(text)
if (text.endsWith('s')) return Number.parseFloat(text) * 1000
return Number.parseFloat(text) || 0
if (text.endsWith('ms')) return Math.round(Number.parseFloat(text))
if (text.endsWith('s')) return Math.round(Number.parseFloat(text) * 1000)
return Math.round(Number.parseFloat(text) || 0)
}
function percentile(sorted, p) {
@@ -97,12 +95,20 @@ function percentile(sorted, p) {
return Math.round(sorted[index])
}
function formatNumber(value) {
const n = Number(value || 0)
if (n >= 1000000) return `${(n / 1000000).toFixed(1)}M`
if (n >= 10000) return `${Math.round(n / 1000)}K`
return n.toLocaleString('zh-CN')
}
async function refresh() {
try {
const nextStatus = await GetStatus()
status.value = nextStatus
emit('status', nextStatus)
requests.value = await GetRequests()
tokenStats.value = await GetTokenStats()
config.value = await GetConfig()
if (nextStatus.running) {
models.value = await GetModels()
@@ -113,7 +119,7 @@ async function refresh() {
}
async function refreshModels() {
loading.value = true
modelsLoading.value = true
try {
models.value = await RefreshModels()
emit('log', 'info', `模型探测完成：${models.value.length} 个`)
@@ -121,7 +127,7 @@ async function refreshModels() {
} catch (e) {
emit('log', 'error', '模型探测失败：' + (e.message || String(e)) + '。请确认 Lingma 插件已启动并登录；自动探测失败时可到设置页手动填写 WebSocket：ws://127.0.0.1:36510/，或 Windows Named Pipe：\\\\.\\pipe\\lingma-xxxx。')
} finally {
loading.value = false
modelsLoading.value = false
}
}
@@ -141,7 +147,7 @@ async function copyModelName(model) {
}
async function toggleProxy() {
loading.value = true
proxyLoading.value = true
try {
if (isRunning.value) {
await StopProxy()
@@ -154,13 +160,13 @@ async function toggleProxy() {
} catch (e) {
emit('log', 'error', '代理切换失败：' + (e.message || String(e)))
} finally {
loading.value = false
proxyLoading.value = false
}
}
async function restartProxy() {
if (!isRunning.value) return
loading.value = true
proxyLoading.value = true
try {
await StopProxy()
await StartProxy()
@@ -169,7 +175,7 @@ async function restartProxy() {
} catch (e) {
emit('log', 'error', '代理重启失败：' + (e.message || String(e)))
} finally {
loading.value = false
proxyLoading.value = false
}
}
@@ -241,9 +247,9 @@ onUnmounted(() => {
<strong>{{ sessionLabel }}</strong>
</div>
<div class="strip-actions">
<button :class="{ active: !isRunning }" type="button" :disabled="loading || isRunning" @click="toggleProxy">启动</button>
<button :class="{ active: isRunning }" type="button" :disabled="loading || !isRunning" @click="toggleProxy">停止</button>
<button type="button" :disabled="loading || !isRunning" @click="restartProxy">重启</button>
<button :class="{ active: !isRunning }" type="button" :disabled="proxyLoading || isRunning" @click="toggleProxy">启动</button>
<button :class="{ active: isRunning }" type="button" :disabled="proxyLoading || !isRunning" @click="toggleProxy">停止</button>
<button type="button" :disabled="proxyLoading || !isRunning" @click="restartProxy">重启</button>
</div>
</section>
@@ -257,12 +263,7 @@ onUnmounted(() => {
<span class="status-chip ok">Healthy</span>
</div>
<div class="activity-chart" aria-label="延迟趋势图">
<span
v-for="(height, index) in chartBars"
:key="index"
class="bar"
:style="{ height: `${height}%`, opacity: 0.55 + index / 45 }"
></span>
<span v-for="(height, index) in chartBars" :key="index" class="bar" :style="{ height: `${height}%`, opacity: 0.55 + index / 45 }"></span>
<span v-if="chartBars.length === 0" class="chart-empty">暂无请求</span>
</div>
<div class="health-stats">
@@ -278,22 +279,13 @@ onUnmounted(() => {
<div>
<h2>Models</h2>
</div>
<button class="secondary-button" type="button" :disabled="loading || !isRunning" @click="refreshModels">探测模型</button>
<button class="btn-sm-outline" type="button" :disabled="modelsLoading || !isRunning" @click="refreshModels">
{{ modelsLoading ? '探测中...' : '探测模型' }}
</button>
</div>
<div class="model-card-list hidden-scrollbar">
<button
v-for="model in displayModels"
:key="model.id"
class="model-row model-choice"
type="button"
:title="`复制模型 ID：${model.id}`"
@click="copyModelName(model)"
>
<span
class="model-brand-icon"
:style="{ '--model-icon': `url(${modelIcon(model).src})`, '--model-icon-color': modelIcon(model).color }"
aria-hidden="true"
></span>
<button v-for="model in displayModels" :key="model.id" class="model-row model-choice" type="button" :title="`复制模型 ID：${model.id}`" @click="copyModelName(model)">
<span class="model-brand-icon" :style="{ '--model-icon': `url(${modelIcon(model).src})`, '--model-icon-color': modelIcon(model).color }" aria-hidden="true"></span>
<div>
<div class="model-name">{{ model.name || model.id }}</div>
</div>
@@ -301,75 +293,84 @@ onUnmounted(() => {
</button>
</div>
<div v-if="displayModels.length === 0" class="empty-state compact">暂无模型，启动代理后点击探测模型。</div>
<button class="link-row" type="button" @click="emit('open-models')">查看全部模型 <i class="bi bi-chevron-right"></i></button>
</div>
<div class="glass-panel area-config">
<div class="panel-header">
<div class="panel-header compact-header">
<div>
<h2>Configuration</h2>
<p>首页只展示关键配置，完整项在设置页查看。</p>
</div>
<span class="status-chip ok">Valid</span>
</div>
<div class="setting-row">
<div class="config-summary">
<div class="config-summary-item">
<label>监听地址</label>
<strong>{{ config.Host || '127.0.0.1' }}:{{ config.Port || 8095 }}</strong>
</div>
<div class="config-summary-item">
<label>传输方式</label>
<strong>{{ transportLabel }}</strong>
</div>
<div class="config-summary-item">
<label>会话策略</label>
<strong>{{ config.SessionMode || 'Reuse' }}</strong>
</div>
<div class="config-summary-item">
<label>超时</label>
<strong>{{ config.Timeout || 120 }} 秒</strong>
</div>
<div class="config-summary-item span-2">
<label>工作目录</label>
<strong :title="config.Cwd || '未配置'">{{ config.Cwd || '未配置' }}</strong>
</div>
<div v-if="config.CurrentFilePath" class="config-summary-item span-2">
<label>当前文件</label>
<strong :title="config.CurrentFilePath">{{ config.CurrentFilePath }}</strong>
</div>
</div>
</div>
<div class="glass-panel area-usage">
<div class="panel-header">
<div>
<div class="cell-main">Host</div>
<div class="cell-sub">{{ config.Host || '127.0.0.1' }}</div>
<h2>Token 统计</h2>
<p>按代理返回的 usage 累计，流式缺失字段时只统计可获得部分。</p>
</div>
<span class="status-chip ok"><i class="bi bi-check"></i></span>
<span class="status-chip ok">Persisted</span>
</div>
<div class="setting-row">
<div class="usage-grid">
<div>
<div class="cell-main">Port</div>
<div class="cell-sub">{{ config.Port || 8095 }}</div>
<label>总 Token</label>
<strong>{{ formatNumber(tokenStats.totalTokens) }}</strong>
</div>
<span class="status-chip ok"><i class="bi bi-check"></i></span>
</div>
<div class="setting-row">
<div>
<div class="cell-main">Transport</div>
<div class="cell-sub">{{ transportLabel }}</div>
<label>输入</label>
<strong>{{ formatNumber(tokenStats.inputTokens) }}</strong>
</div>
<span class="status-chip ok"><i class="bi bi-check"></i></span>
</div>
<div class="setting-row">
<div>
<div class="cell-main">Session</div>
<div class="cell-sub">{{ config.SessionMode || 'Reuse' }}</div>
<label>输出</label>
<strong>{{ formatNumber(tokenStats.outputTokens) }}</strong>
</div>
<span class="status-chip ok"><i class="bi bi-check"></i></span>
</div>
<div class="setting-row">
<div>
<div class="cell-main">Timeout (s)</div>
<div class="cell-sub">{{ config.Timeout || 120 }} 秒</div>
<label>成功率</label>
<strong>{{ successRate }}</strong>
</div>
<span class="status-chip ok"><i class="bi bi-check"></i></span>
</div>
<div class="setting-row">
<div>
<div class="cell-main">CWD</div>
<div class="cell-sub">{{ config.Cwd || '未配置' }}</div>
</div>
<span class="status-chip ok"><i class="bi bi-check"></i></span>
</div>
<div class="setting-row">
<div>
<div class="cell-main">Current File</div>
<div class="cell-sub">{{ config.CurrentFilePath || '未配置' }}</div>
</div>
<span class="status-chip ok"><i class="bi bi-check"></i></span>
<div class="usage-foot">
<span>累计请求 {{ formatNumber(tokenStats.totalRequests) }} 次</span>
<span v-if="tokenStats.lastModel">最近模型 {{ tokenStats.lastModel }}</span>
</div>
</div>
<div class="table-panel area-requests">
<div class="table-toolbar">
<div>
<div class="panel-header" style="margin: 0">
<div class="panel-header toolbar-header">
<h2>Recent Requests</h2>
</div>
</div>
<button class="secondary-button" type="button" @click="emit('open-requests')">查看全部</button>
<button type="button" class="btn-sm-outline" @click="emit('open-requests')">
查看全部请求 <i class="bi bi-chevron-right"></i>
</button>
</div>
<div v-if="displayRequests.length > 0" class="table-scroll hidden-scrollbar">
<table class="data-table">
@@ -398,10 +399,6 @@ onUnmounted(() => {
</table>
</div>
<div v-else class="empty-state compact">暂无请求记录。连接客户端后会显示真实调用。</div>
<div class="table-footer">
<span>Showing {{ displayRequests.length }} of {{ requests.length }}</span>
<button type="button" @click="emit('open-requests')">查看全部请求 <i class="bi bi-chevron-right"></i></button>
</div>
</div>
</section>
</div>

83
desktop/frontend/src/views/Models.vue
View File

@@ -17,20 +17,81 @@ const filtered = computed(() => {
return models.value.filter((model) => `${model.id} ${model.name}`.toLowerCase().includes(q))
})
function modelTag(model) {
function modelSpec(model) {
const text = `${model.id} ${model.name}`.toLowerCase()
if (text.includes('coder')) return '工具优先'
if (text.includes('thinking')) return '推理'
if (text.includes('kimi')) return '长文本'
if (text.includes('minimax')) return '通用'
return 'Lingma'
if (text.includes('kmodel') || text.includes('kimi')) {
return {
context: '256K',
capability: '文本/图像/视频/工具',
source: 'Kimi 官方',
}
}
if (text.includes('mmodel') || text.includes('minimax')) {
return {
context: '200K',
capability: 'Agent / Tool Use',
source: 'MiniMax 官方',
}
}
if (text.includes('coder')) {
return {
context: '1M',
capability: '思考 / Function Calling / 结构化输出',
source: '阿里云百炼 Qwen3-Coder',
}
}
if (text.includes('thinking')) {
return {
context: '256K',
capability: '思考 / Function Calling / 推理',
source: '阿里云百炼 Qwen3',
}
}
if (text.includes('qwen_max') || text.includes('qwen3-max')) {
return {
context: '256K',
capability: '思考 / Function Calling / 内置工具',
source: '阿里云百炼 Qwen3-Max',
}
}
if (text.includes('qmodel') || text.includes('qwen3.6')) {
return {
context: '1M',
capability: 'Function Calling / 内置工具 / 结构化输出',
source: '阿里云百炼 Qwen3.6-Plus',
}
}
if (text.includes('auto')) {
return {
context: '自动',
capability: 'Lingma 自动路由',
source: '账号返回',
}
}
return {
context: '未公开',
capability: '通用',
source: '账号返回',
}
}
async function loadCachedModels() {
loading.value = true
try {
status.value = await GetStatus()
models.value = await GetModels()
} catch (e) {
emit('log', 'error', '模型缓存读取失败：' + (e.message || String(e)))
} finally {
loading.value = false
}
}
async function refresh() {
loading.value = true
try {
status.value = await GetStatus()
models.value = status.value.running ? await RefreshModels() : await GetModels()
models.value = await RefreshModels()
emit('log', 'info', `模型列表刷新完成：${models.value.length} 个`)
} catch (e) {
emit('log', 'error', '模型列表刷新失败：' + (e.message || String(e)) + '。自动探测失败时请到设置页手动填写 WebSocket：ws://127.0.0.1:36510/，或 Windows Named Pipe：\\\\.\\pipe\\lingma-xxxx。')
@@ -54,7 +115,7 @@ async function copyModelName(model) {
}
}
onMounted(refresh)
onMounted(loadCachedModels)
</script>
<template>
@@ -111,7 +172,11 @@ onMounted(refresh)
<div class="model-name">{{ model.name || model.id }}</div>
<div class="model-meta">{{ model.id }}</div>
</div>
<span class="status-chip" :class="modelTag(model) === '工具优先' ? 'ok' : 'warn'">{{ modelTag(model) }}</span>
<div class="model-specs">
<span class="spec-chip strong">{{ modelSpec(model).context }}</span>
<span class="spec-chip">{{ modelSpec(model).capability }}</span>
<span class="spec-chip muted-chip">{{ modelSpec(model).source }}</span>
</div>
</button>
</div>
<div v-else class="empty-state">启动代理并刷新后会显示模型。</div>

9
desktop/frontend/src/views/Requests.vue
View File

@@ -14,7 +14,7 @@ const activeStatus = ref('all')
const filtered = computed(() => {
const q = query.value.trim().toLowerCase()
return requests.value.filter((request) => {
const matchesQuery = !q || `${request.method} ${request.path} ${request.statusCode}`.toLowerCase().includes(q)
const matchesQuery = !q || `${request.method} ${request.path} ${request.statusCode} ${request.model || ''}`.toLowerCase().includes(q)
const code = Number(request.statusCode)
const matchesStatus =
activeStatus.value === 'all' ||
@@ -117,7 +117,10 @@ onUnmounted(() => {
<section class="table-panel requests-panel">
<div class="table-toolbar">
<input v-model="query" class="search-input" type="search" placeholder="搜索路径、方法或状态码" />
<div class="toolbar-search-wrap">
<input v-model="query" class="search-input toolbar-search-input" type="search" placeholder="搜索路径、方法或状态码" />
<span class="muted toolbar-count">Showing {{ filtered.length }} of {{ requests.length }}</span>
</div>
<div class="segmented">
<button :class="{ active: activeStatus === 'all' }" type="button" @click="activeStatus = 'all'">全部</button>
<button :class="{ active: activeStatus === 'ok' }" type="button" @click="activeStatus = 'ok'">成功</button>
@@ -133,6 +136,7 @@ onUnmounted(() => {
<th>时间</th>
<th>方法</th>
<th>路径</th>
<th>模型</th>
<th>状态</th>
<th>耗时</th>
</tr>
@@ -150,6 +154,7 @@ onUnmounted(() => {
<div class="cell-main">{{ request.path }}</div>
<div class="cell-sub">{{ request.reqBody ? '包含请求体' : '无请求体' }}</div>
</td>
<td>{{ request.model || '-' }}</td>
<td><span class="status-chip" :class="statusClass(request.statusCode)">{{ request.statusCode }}</span></td>
<td>{{ request.duration }}</td>
</tr>

41
desktop/frontend/src/views/Settings.vue
View File

@@ -8,6 +8,8 @@ const config = ref({})
const detection = ref(null)
const saving = ref(false)
const openSelect = ref('')
const fallbackModelsText = ref('')
const isIPCBackend = computed(() => (config.value.Backend || 'ipc') === 'ipc')
const selectOptions = {
Backend: [
@@ -54,6 +56,9 @@ function chooseOption(field, value) {
onMounted(async () => {
try {
config.value = await GetConfig()
fallbackModelsText.value = Array.isArray(config.value.RemoteFallbackModels)
? config.value.RemoteFallbackModels.join('\n')
: ''
await refreshDetection()
} catch (e) {
emit('log', 'error', '配置加载失败：' + (e.message || String(e)))
@@ -71,6 +76,10 @@ async function refreshDetection() {
async function save() {
saving.value = true
try {
config.value.RemoteFallbackModels = fallbackModelsText.value
.split(/\n|,/)
.map((item) => item.trim())
.filter(Boolean)
await UpdateConfig(config.value)
await refreshDetection()
emit('log', 'info', '配置已保存，代理已按需重启')
@@ -95,7 +104,7 @@ async function save() {
</button>
</div>
<section class="grid-2">
<section class="grid-2 settings-grid">
<div class="glass-panel">
<div class="panel-header">
<div>
@@ -156,6 +165,23 @@ async function save() {
<label>超时秒数</label>
<input v-model.number="config.Timeout" type="number" min="1" />
</div>
<div class="field span-2 switch-field">
<div>
<label>远端超时兜底</label>
<p>远端 API 超时、限流或 5xx 且尚未流式输出时，自动切换到下一个可用模型。</p>
</div>
<label class="switch">
<input v-model="config.RemoteFallbackEnabled" type="checkbox" />
<span></span>
</label>
</div>
<div class="field span-2">
<label>兜底模型顺序</label>
<textarea
v-model="fallbackModelsText"
placeholder="kmodel&#10;mmodel&#10;dashscope_qwen3_coder&#10;dashscope_qmodel"
></textarea>
</div>
<div class="field span-2">
<label>WebSocket 地址</label>
<input v-model="config.WebSocketURL" type="text" placeholder="留空自动探测 Lingma WebSocket" />
@@ -231,10 +257,16 @@ async function save() {
<div class="panel-header">
<div>
<h2>会话与环境</h2>
<p>影响 Lingma 会话上下文和工具执行环境。</p>
<p>仅在 IPC 插件模式下生效，影响 Lingma 会话上下文和工具执行环境。</p>
</div>
<span class="status-chip" :class="isIPCBackend ? 'ok' : 'warn'">{{ isIPCBackend ? '仅 IPC 生效' : '远端模式忽略' }}</span>
</div>
<div class="form-grid">
<div v-if="!isIPCBackend" class="hint-box compact-hint">
<strong>当前为远端 API 模式</strong>
<span>右侧这组参数不会参与远端请求，只在切换到 IPC 插件模式后生效。</span>
</div>
<fieldset class="settings-fieldset" :disabled="!isIPCBackend">
<div class="form-grid compact-form-grid">
<div class="field">
<label>模式</label>
<div class="custom-select" :class="{ open: openSelect === 'Mode' }">
@@ -301,9 +333,10 @@ async function save() {
</div>
<div class="field span-2">
<label>工作目录</label>
<textarea v-model="config.Cwd" placeholder="Lingma 创建 session 时使用的 cwd"></textarea>
<input v-model="config.Cwd" type="text" placeholder="Lingma 创建 session 时使用的 cwd" />
</div>
</div>
</fieldset>
</div>
</section>
</div>

4
desktop/frontend/wailsjs/go/main/App.d.ts vendored
View File

@@ -11,12 +11,16 @@ export function GetConfig():Promise<service.Config>;
export function GetDetectionInfo():Promise<main.DetectionInfo>;
export function GetLogs():Promise<Array<main.AppLog>>;
export function GetModels():Promise<Array<main.ModelInfo>>;
export function GetRequests():Promise<Array<main.RequestRecord>>;
export function GetStatus():Promise<main.ProxyStatus>;
export function GetTokenStats():Promise<main.TokenStats>;
export function HideWindow():Promise<void>;
export function MinimizeWindow():Promise<void>;

8
desktop/frontend/wailsjs/go/main/App.js
View File

@@ -18,6 +18,10 @@ export function GetDetectionInfo() {
return window['go']['main']['App']['GetDetectionInfo']();
}
export function GetLogs() {
return window['go']['main']['App']['GetLogs']();
}
export function GetModels() {
return window['go']['main']['App']['GetModels']();
}
@@ -30,6 +34,10 @@ export function GetStatus() {
return window['go']['main']['App']['GetStatus']();
}
export function GetTokenStats() {
return window['go']['main']['App']['GetTokenStats']();
}
export function HideWindow() {
return window['go']['main']['App']['HideWindow']();
}

52
desktop/frontend/wailsjs/go/models.ts
View File

@@ -1,5 +1,21 @@
export namespace main {
export class AppLog {
time: string;
level: string;
message: string;
static createFrom(source: any = {}) {
return new AppLog(source);
}
constructor(source: any = {}) {
if ('string' === typeof source) source = JSON.parse(source);
this.time = source["time"];
this.level = source["level"];
this.message = source["message"];
}
}
export class DetectionInfo {
listenUrl: string;
backend: string;
@@ -86,6 +102,9 @@ export namespace main {
statusCode: number;
duration: string;
size?: string;
inputTokens?: number;
outputTokens?: number;
totalTokens?: number;
reqBody?: string;
respBody?: string;
@@ -102,10 +121,39 @@ export namespace main {
this.statusCode = source["statusCode"];
this.duration = source["duration"];
this.size = source["size"];
this.inputTokens = source["inputTokens"];
this.outputTokens = source["outputTokens"];
this.totalTokens = source["totalTokens"];
this.reqBody = source["reqBody"];
this.respBody = source["respBody"];
}
}
export class TokenStats {
totalRequests: number;
successRequests: number;
inputTokens: number;
outputTokens: number;
totalTokens: number;
byModel?: Record<string, number>;
lastModel?: string;
lastUpdated?: string;
static createFrom(source: any = {}) {
return new TokenStats(source);
}
constructor(source: any = {}) {
if ('string' === typeof source) source = JSON.parse(source);
this.totalRequests = source["totalRequests"];
this.successRequests = source["successRequests"];
this.inputTokens = source["inputTokens"];
this.outputTokens = source["outputTokens"];
this.totalTokens = source["totalTokens"];
this.byModel = source["byModel"];
this.lastModel = source["lastModel"];
this.lastUpdated = source["lastUpdated"];
}
}
}
@@ -128,6 +176,8 @@ export namespace service {
ShellType: string;
SessionMode: string;
Timeout: number;
RemoteFallbackEnabled: boolean;
RemoteFallbackModels: string[];
static createFrom(source: any = {}) {
return new Config(source);
@@ -151,6 +201,8 @@ export namespace service {
this.ShellType = source["ShellType"];
this.SessionMode = source["SessionMode"];
this.Timeout = source["Timeout"];
this.RemoteFallbackEnabled = source["RemoteFallbackEnabled"];
this.RemoteFallbackModels = source["RemoteFallbackModels"];
}
}

6
desktop/main.go
View File

@@ -2,6 +2,7 @@ package main
import (
"embed"
"os"
goruntime "runtime"
"github.com/wailsapp/wails/v2"
@@ -17,6 +18,7 @@ var assets embed.FS
func main() {
app := NewApp()
enableInspector := os.Getenv("LINGMA_DESKTOP_DEBUG") == "1"
err := wails.Run(&options.App{
Title: "Lingma IPC Proxy",
@@ -28,6 +30,10 @@ func main() {
AssetServer: &assetserver.Options{
Assets: assets,
},
EnableDefaultContextMenu: enableInspector,
Debug: options.Debug{
OpenInspectorOnStartup: enableInspector,
},
BackgroundColour: &options.RGBA{R: 15, G: 23, B: 42, A: 1},
Menu: appMenu(app),
OnStartup: app.startup,

2
desktop/wails.json
View File

@@ -11,6 +11,6 @@
"email": "lutc5@asiainfo.com"
},
"info": {
"productVersion": "1.4.2"
"productVersion": "1.4.3"
}
}

612
docs/architecture.md
View File

@@ -1,424 +1,316 @@
# lingma-ipc-proxy 架构文档
# lingma-ipc-proxy Architecture
本文档描述 lingma-ipc-proxy 的系统架构、工作原理和核心流程。
This document describes the current architecture of `lingma-ipc-proxy`, including both backend modes:
- `ipc`: bridge to the local Lingma IDE plugin transport
- `remote`: call Lingma remote HTTP APIs directly with detected credentials
---
## 1. 整体架构
## 1. System Overview
```
┌─────────────────────────────────────────────────────────────────────────┐
│ 客户端层 │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ │
│ │ Claude Code │ │ OpenAI │ │ Cline │ │ Continue │ │
│ │ (Anthropic) │ │ SDK │ │ (OpenAI) │ │ (OpenAI) │ │
│ └──────┬───────┘ └──────┬───────┘ └──────┬───────┘ └──────┬───────┘ │
└─────────┼─────────────────┼─────────────────┼─────────────────┼─────────┘
│ │ │ │
└─────────────────┴────────┬────────┴─────────────────┘
│ HTTP
▼
┌─────────────────────────────────────────────────────────────────────────┐
│ lingma-ipc-proxy │
│ ┌─────────────────────────────────────────────────────────────────┐ │
│ │ internal/httpapi │ │
│ │ ┌─────────────┐ ┌─────────────┐ ┌─────────────────────────┐ │ │
│ │ │ /v1/models │ │/v1/chat/comp│ │ /v1/messages │ │ │
│ │ │ (GET) │ │ (POST) │ │ (POST) │ │ │
│ │ └──────┬──────┘ └──────┬──────┘ └───────────┬─────────────┘ │ │
│ │ └─────────────────┴──────────┬──────────┘ │ │
│ │ │ normalizeRequest │ │
│ │ ▼ │ │
│ │ ┌─────────────────────────────────────────────────────────┐ │ │
│ │ │ internal/service │ │ │
│ │ │ ┌──────────┐ ┌──────────┐ ┌────────────────────────┐ │ │ │
│ │ │ │ Session │ │ Prompt │ │ Stream/Event │ │ │ │
│ │ │ │ Manager │ │ Builder │ │ Handler │ │ │ │
│ │ │ └────┬─────┘ └────┬─────┘ └───────────┬────────────┘ │ │ │
│ │ │ └─────────────┴──────────┬─────────┘ │ │ │
│ │ │ │ buildLingmaPrompt │ │ │
│ │ │ ▼ │ │ │
│ │ │ ┌─────────────────────────────────────────────────┐ │ │ │
│ │ │ │ internal/lingmaipc │ │ │ │
│ │ │ │ ┌──────────────┐ ┌──────────────────────────┐ │ │ │ │
│ │ │ │ │ WebSocket │ │ Named Pipe (Win) │ │ │ │ │
│ │ │ │ │ Transport │ │ Transport │ │ │ │ │
│ │ │ │ └──────┬───────┘ └───────────┬──────────────┘ │ │ │ │
│ │ │ └─────────┼──────────────────────┼────────────────┘ │ │ │
│ │ └────────────┼──────────────────────┼────────────────────┘ │ │
│ │ │ │ │ │
│ │ ┌────────────┼──────────────────────┼────────────────────┐ │ │
│ │ │ ▼ ▼ │ │ │
│ │ │ ┌─────────────────────────────────────────────────┐ │ │ │
│ │ │ │ internal/toolemulation │ │ │ │
│ │ │ │ ┌──────────────┐ ┌──────────────────────────┐ │ │ │ │
│ │ │ │ │InjectTooling │ │ ParseActionBlocks │ │ │ │ │
│ │ │ │ │ (Prompt) │ │ (Response) │ │ │ │ │
│ │ │ │ └──────────────┘ └──────────────────────────┘ │ │ │ │
│ │ │ └─────────────────────────────────────────────────┘ │ │ │
│ │ └───────────────────────────────────────────────────────┘ │ │
│ └───────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────────┘
│
│ WebSocket / Named Pipe
▼
┌─────────────────────────────────────────────────────────────────────────┐
│ Lingma 后端进程 │
│ (VS Code 插件的本地 IPC 服务) │
│ ws://127.0.0.1:8899/ws │
└─────────────────────────────────────────────────────────────────────────┘
│
│ HTTP API
▼
┌─────────────────────────────────────────────────────────────────────────┐
│ 云端模型服务 │
│ (Kimi-K2.6 / Qwen3-Max / MiniMax-M2.7 等) │
└─────────────────────────────────────────────────────────────────────────┘
```mermaid
flowchart LR
A["Clients<br/>Claude Code / Hermes / Cline / Continue / OpenAI SDK / Anthropic SDK"]
B["internal/httpapi<br/>OpenAI + Anthropic compatible routes"]
C["internal/service<br/>request normalization / session policy / streaming / fallback"]
D["internal/toolemulation<br/>tool prompt injection + action block parsing"]
E["internal/lingmaipc<br/>WebSocket / Named Pipe"]
F["internal/remote<br/>credential detection / model list / chat / SSE"]
G["Lingma plugin local process"]
H["Lingma remote API"]
I["Desktop app<br/>Wails GUI / logs / token stats / persisted state"]
A --> B
I --> B
B --> C
C --> D
C --> E
C --> F
E --> G
F --> H
```
---
## 2. 模块职责
## 2. Runtime Modes
### 2.1 internal/httpapi
### 2.1 IPC mode
HTTP API 适配层，负责将外部请求转换为内部 `service.ChatRequest`。
`backend=ipc`
| 端点 | 协议 | 功能 |
|------|------|------|
| `GET /v1/models` | OpenAI | 返回可用模型列表 |
| `POST /v1/chat/completions` | OpenAI | 聊天补全（流式/非流式） |
| `POST /v1/messages` | Anthropic | 消息接口（流式/非流式） |
- Reads local plugin transport information
- Connects through:
- WebSocket on macOS / Linux
- Named Pipe on Windows
- Reuses Lingma plugin session semantics
- Session/environment options in the desktop UI apply only here
**核心函数：**
- `handleOpenAIChatCompletions()` - 处理 OpenAI 格式请求
- `handleAnthropicMessages()` - 处理 Anthropic 格式请求
- `normalizeOpenAIRequest()` / `normalizeAnthropicRequest()` - 归一化请求
### 2.2 Remote API mode
**关键设计：**
- 支持 CORS 预检请求 (`OPTIONS`)
- 单请求并发控制 (`tryAcquire()` / `release()`)
- 流式响应通过 `http.Flusher` 实现 SSE
`backend=remote`
### 2.2 internal/service
业务逻辑层，负责会话管理和 Prompt 构建。
**核心结构：**
```go
type Service struct {
cfg Config
client *lingmaipc.Client
stickySessionID string
stickyModelID string
}
```
**核心函数：**
- `Generate()` - 非流式生成
- `GenerateStream()` - 流式生成（返回 `events` + `done` channel）
- `buildLingmaPrompt()` - 构建 Lingma 原生 Prompt
- `runPromptLocked()` - 发送 `session/prompt` RPC 并监听 `session/update` 通知
**会话模式：**
| 模式 | 行为 |
|------|------|
| `reuse` | 复用 sticky session，多轮对话保持上下文 |
| `fresh` | 每个请求新建临时 session，完成后删除 |
| `auto` | 单轮请求复用；带 system/history 的请求用 fresh |
### 2.3 internal/lingmaipc
IPC 通信层，负责与 Lingma 后端进程建立连接。
**传输方式：**
| 平台 | 默认传输 | 说明 |
|------|----------|------|
| Windows | Named Pipe | `\\.\pipe\lingma-*` |
| macOS/Linux | WebSocket | `ws://127.0.0.1:{port}/ws` |
**连接发现：**
- 读取 VS Code 插件缓存：`~/.config/Lingma/SharedClientCache/.info.json`
- 获取 WebSocket 端口号
- 自动重连机制
**RPC 协议：**
- `session/new` - 创建会话
- `session/prompt` - 发送用户消息
- `session/update` - 接收流式响应通知
- `session/set_model` - 切换模型
- `chat/deleteSessionById` - 删除会话
### 2.4 internal/toolemulation
Tool 调用模拟层，将标准 `tools` 协议转换为 Prompt 层契约。
**核心流程：**
```
Client tools ──→ ExtractAnthropicTools() ──→ []Tool
│
▼
InjectTooling() ──→ System Prompt + Tool 说明
│
▼
模型输出 action block
│
▼
ParseActionBlocks() ──→ []ToolCall
│
▼
编码为 Anthropic tool_use / OpenAI tool_calls
```
**Prompt 契约格式：**
```
```json action
{"tool":"NAME","parameters":{"key":"value"}}
```
```
**支持格式：**
- `{"tool":"X","parameters":{}}` ✅ 标准格式
- `{"tool":"X","arguments":{}}` ✅ 兼容格式
- `{"tool":"X","input":{}}` ✅ 兼容格式
- `{"tool":"X","arg1":"val"}` ✅ 顶层参数（部分模型）
- Reads Lingma remote base URL
- Loads credentials from:
- explicit `remote_auth_file`
- or detected Lingma cache under `~/.lingma`
- Calls remote model list and chat endpoints directly
- Supports timeout / 429 / 5xx fallback across available remote models
- Does not use local plugin session environment knobs
---
## 3. 核心流程
## 3. Module Responsibilities
### 3.1 普通聊天请求流程
### 3.1 `cmd/lingma-ipc-proxy`
Entry point and config loading.
Responsibilities:
- parse CLI flags
- merge config file + environment + flags
- choose backend mode
- build `service.Config`
- start `internal/httpapi.Server`
Important config fields:
- `backend`
- `transport`
- `websocket_url`
- `pipe`
- `remote_base_url`
- `remote_auth_file`
- `remote_version`
- `remote_fallback_enabled`
- `remote_fallback_models`
### 3.2 `internal/httpapi`
Compatibility layer for OpenAI and Anthropic style APIs.
Primary routes:
- `GET /v1/models`
- `POST /v1/chat/completions`
- `POST /v1/messages`
- `GET /health`
- `GET /props`
Responsibilities:
- normalize OpenAI / Anthropic requests into `service.ChatRequest`
- convert service results back to OpenAI / Anthropic payloads
- stream SSE responses
- sanitize and record request / response payloads for debug UI
### 3.3 `internal/service`
Core orchestration layer.
Responsibilities:
- choose active backend
- warm up backend connection / credentials
- list models
- generate non-streaming responses
- generate streaming responses
- apply session reuse policy in IPC mode
- inject / parse tool emulation
- normalize image inputs
- apply remote fallback order
Important behavior split:
- IPC path uses `internal/lingmaipc`
- Remote path uses `internal/remote`
### 3.4 `internal/lingmaipc`
Local transport client for Lingma plugin IPC.
Responsibilities:
- detect WebSocket / pipe endpoint
- dial and reconnect
- send RPC messages such as:
- `session/new`
- `session/prompt`
- `session/set_model`
- `chat/deleteSessionById`
- consume `session/update` notifications
### 3.5 `internal/remote`
Remote HTTP client for Lingma cloud APIs.
Responsibilities:
- resolve base URL
- load and validate credentials
- derive machine / user identity for remote auth
- list remote models
- call remote chat endpoint
- handle remote SSE streaming
### 3.6 `internal/toolemulation`
Prompt-based tool bridge for models that do not expose native tool calling in Lingma transport.
Responsibilities:
- extract tool definitions from OpenAI / Anthropic requests
- append tool contract to prompt
- parse JSON action blocks from model output
- project tool calls back to:
- Anthropic `tool_use`
- OpenAI `tool_calls`
---
## 4. Request Flow
### 4.1 Shared ingress flow
```mermaid
sequenceDiagram
participant C as Client
participant H as HTTP API
participant S as Service
participant L as Lingma IPC
participant B as Lingma Backend
participant Client
participant HTTP as httpapi
participant Service as service
C->>H: POST /v1/messages
H->>H: normalizeAnthropicRequest()
H->>S: GenerateStream(req)
S->>S: ensureConnected()
S->>S: resolveSession()
S->>S: buildLingmaPrompt()
S->>L: Send("session/prompt", params)
L->>B: WebSocket RPC
B->>L: session/update (agent_message_chunk)
loop 流式响应
L->>S: notification (chunk)
S->>H: events <- StreamEvent{Delta}
H->>C: SSE: content_block_delta
end
B->>L: session/update (chat_finish)
L->>S: notification (finish)
S->>H: done <- StreamResult
H->>C: SSE: message_stop
Client->>HTTP: OpenAI/Anthropic request
HTTP->>HTTP: normalize request
HTTP->>Service: Generate / GenerateStream
```
### 3.2 Tool 调用流程
### 4.2 IPC backend flow
```mermaid
sequenceDiagram
participant C as Client
participant H as HTTP API
participant T as ToolEmulation
participant S as Service
participant L as Lingma IPC
participant Service
participant Tool as toolemulation
participant IPC as lingmaipc
participant Plugin as Lingma plugin
C->>H: POST /v1/messages (with tools)
H->>T: ExtractAnthropicTools()
H->>S: GenerateStream(req)
S->>T: InjectTooling(system, tools)
S->>L: session/prompt (with tool prompt)
L->>S: response (with action blocks)
S->>T: ParseActionBlocks(text)
T->>S: []ToolCall
S->>H: ChatResult{Text, ToolCalls}
H->>C: SSE: tool_use blocks
C->>H: POST /v1/messages (tool_result)
H->>T: ActionOutputPrompt(toolUseID, content)
H->>S: GenerateStream(req)
S->>L: session/prompt (with tool result)
L->>S: response
S->>H: ChatResult
H->>C: SSE: final response
Service->>Tool: inject tool contract if needed
Service->>IPC: ensure connected
Service->>IPC: create/reuse session
Service->>IPC: session/prompt
IPC->>Plugin: RPC message
Plugin-->>IPC: session/update chunks
IPC-->>Service: stream events
Service-->>Service: parse tool blocks / image references / stop reason
```
### 3.3 图片传输流程
### 4.3 Remote backend flow
```mermaid
sequenceDiagram
participant C as Client
participant H as HTTP API
participant S as Service
participant L as Lingma IPC
participant Service
participant Remote as remote client
participant API as Lingma remote API
C->>H: POST /v1/messages (with image)
H->>H: extractAnthropicImages()
H->>S: ChatRequest{Images: [...]}
S->>S: runPromptLocked()
Note over S: 1. 保存 base64 到 /tmp/lingma-img-*.ext
Note over S: 2. 构建 URI: lingma:///agent/file?path=...
S->>L: session/prompt
Note over L: prompt: [{type:"text"}, {type:"image", mimeType, uri, data}]
L->>S: response (model sees image)
S->>H: ChatResult
H->>C: SSE response
```
### 3.4 流式输出 SSE 事件序列
**Anthropic 格式（流式）：**
```
event: message_start
data: {"type":"message_start","message":{...}}
event: content_block_start
data: {"type":"content_block_start","index":0,"content_block":{"type":"text","text":""}}
event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"你"}}
event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"好"}}
... (更多 delta)
event: content_block_stop
data: {"type":"content_block_stop","index":0}
[如有 tool_calls]
event: content_block_start
data: {"type":"content_block_start","index":1,"content_block":{"type":"tool_use","id":"...","name":"Bash","input":{"command":"ls /"}}}
event: content_block_stop
data: {"type":"content_block_stop","index":1}
event: message_delta
data: {"type":"message_delta","delta":{"stop_reason":"end_turn"},"usage":{"output_tokens":5}}
event: message_stop
data: {"type":"message_stop"}
Service->>Remote: load credentials / ensure client
Service->>Remote: list models if needed
Service->>Remote: chat request
Remote->>API: HTTPS request
API-->>Remote: JSON or SSE response
Remote-->>Service: normalized result
Service-->>Service: fallback to next model when allowed
```
---
## 4. 关键技术决策
## 5. Remote Fallback Strategy
### 4.1 为什么使用 Tool Emulation 而非原生 Tool Calling？
Remote fallback is used only when all conditions are true:
Lingma 后端模型（Kimi、Qwen 等）不原生支持 OpenAI/Anthropic 的 `tools` 协议。因此代理层需要将工具定义注入到 Prompt 中，通过结构化文本输出模拟工具调用。
- `backend=remote`
- `remote_fallback_enabled=true`
- request has not emitted stream output yet
- upstream error matches timeout / 429 / 5xx class
**优点：**
- 不依赖上游模型能力
- 兼容任何纯聊天模型
- 可精确控制 Prompt 格式
Current default order:
**缺点：**
- 模型需要学习特定格式
- 解析可能有容错问题
- 增加了 Prompt 长度
1. `kmodel`
2. `mmodel`
3. `dashscope_qwen3_coder`
4. `dashscope_qmodel`
5. `dashscope_qwen_max_latest`
6. `dashscope_qwen_plus_20250428_thinking`
### 4.2 为什么使用 WebSocket/Named Pipe 而非 HTTP？
Lingma 插件使用本地 IPC 与后端通信，优势：
- 低延迟（本地通信）
- 双向实时通知（session/update）
- 认证信息由插件管理，代理无需处理
### 4.3 图片传输的双保险策略
```
Prompt 数组 (Lingma 原生格式):
[
{"type":"text","text":"..."},
{"type":"image","mimeType":"image/png","uri":"lingma:///agent/file?path=...","data":"base64..."}
]
```
- `uri`: Lingma 后端必须验证的本地文件路径
- `data`: base64 编码的图像数据（备用）
- `mimeType`: 图像类型标识
### 4.4 单请求并发控制
Lingma IPC 一次只能处理一个请求，因此代理使用 `tryAcquire()` 机制：
```go
if !s.tryAcquire() {
writeAnthropicError(w, 429, "rate_limit_error",
"Lingma IPC proxy handles one request at a time.")
return
}
defer s.release()
```
Before using that order, the service filters candidates against the actual `/v1/models` result from the remote backend so unavailable models are skipped.
---
## 5. 配置说明
## 6. Desktop App Architecture
### 5.1 配置文件结构
The Wails desktop app is a management UI around the local proxy process.
```json
{
"host": "127.0.0.1",
"port": 8095,
"transport": "websocket",
"mode": "agent",
"shell_type": "zsh",
"session_mode": "auto",
"timeout": 120,
"cwd": "/Users/tiancheng"
}
```
Responsibilities:
### 5.2 配置项说明
- start / stop / restart proxy
- show current backend and resolved endpoints
- persist:
- request history
- logs
- token statistics
- show detected IPC and remote credentials metadata
- edit config and restart proxy on save
| 配置项 | 类型 | 默认值 | 说明 |
|--------|------|--------|------|
| `host` | string | `127.0.0.1` | HTTP 监听地址 |
| `port` | int | `8095` | HTTP 监听端口 |
| `transport` | string | `auto` | IPC 传输方式：`auto`/`pipe`/`websocket` |
| `mode` | string | `chat` | 模式：`chat`/`agent` |
| `shell_type` | string | `powershell` | 终端类型 |
| `session_mode` | string | `auto` | 会话模式：`reuse`/`fresh`/`auto` |
| `timeout` | int | `120` | 请求超时（秒） |
| `cwd` | string | `""` | 工作目录（传给 Lingma 后端） |
Persisted local state:
- config: `~/.config/lingma-ipc-proxy/config.json`
- UI/runtime state: `~/.config/lingma-ipc-proxy/app-state.json`
Production packaging rules:
- packaged app should not auto-open inspector
- local development can opt in with `LINGMA_DESKTOP_DEBUG=1`
---
## 6. 扩展点
## 7. Key Design Decisions
### 6.1 添加新模型
### 7.1 Why keep both IPC and remote modes?
在 `service.go` 的模型映射中添加：
Because the two modes solve different problems:
```go
func (s *Service) resolveInternalModelID(model string) string {
switch strings.ToLower(strings.TrimSpace(model)) {
case "kimi-k2.6":
return "kimi2.6"
case "qwen3-max":
return "qwen3max"
// 添加新模型映射
default:
return ""
}
}
```
- IPC mode preserves plugin session semantics and local tool environment
- Remote mode avoids plugin runtime coupling and is usually better for third-party agent clients
### 6.2 添加新 Tool 格式支持
### 7.2 Why keep tool emulation even with remote mode?
在 `toolemulation.go` 的 `parseToolCallJSON()` 中扩展参数解析逻辑。
Because Lingma-exposed models are not guaranteed to speak OpenAI/Anthropic native tool protocol consistently across all routes. The proxy must keep a stable external contract even when the upstream model capability is uneven.
### 6.3 添加新 API 端点
### 7.3 Why persist requests and token stats in the desktop app?
在 `httpapi/server.go` 的 `NewServer()` 中注册新路由。
Because the GUI is used as an operational console, not a transient preview. Users need model usage, logs, and recent traffic to survive app restarts.
---
*文档版本: 2025-04-25*
*对应代码版本: 当前 master*
## 8. Known Boundaries
- IPC mode still has stronger environment coupling with the local Lingma plugin
- remote credential detection depends on local Lingma cache / auth file layout
- image payloads are sanitized in persisted request logs to avoid oversized local state
- request history may contain mixed models in remote mode when fallback is triggered or when different clients specify different models
---
## 9. Files to Read First
If you are extending the system, start here:
- `cmd/lingma-ipc-proxy/main.go`
- `internal/httpapi/server.go`
- `internal/service/service.go`
- `internal/lingmaipc/*`
- `internal/remote/*`
- `desktop/app.go`
- `desktop/main.go`
---
Document version: 2026-04-30

609
docs/architecture.zh-CN.md
View File

@@ -1,424 +1,315 @@
# lingma-ipc-proxy 架构文档
本文档描述 lingma-ipc-proxy 的系统架构、工作原理和核心流程。
本文档描述 `lingma-ipc-proxy` 的当前架构，覆盖两种后端模式：
- `ipc`：桥接本地 Lingma IDE 插件传输层
- `remote`：直接调用 Lingma 远端 HTTP API
---
## 1. 整体架构
## 1. 系统总览
```
┌─────────────────────────────────────────────────────────────────────────┐
│ 客户端层 │
│ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ │
│ │ Claude Code │ │ OpenAI │ │ Cline │ │ Continue │ │
│ │ (Anthropic) │ │ SDK │ │ (OpenAI) │ │ (OpenAI) │ │
│ └──────┬───────┘ └──────┬───────┘ └──────┬───────┘ └──────┬───────┘ │
└─────────┼─────────────────┼─────────────────┼─────────────────┼─────────┘
│ │ │ │
└─────────────────┴────────┬────────┴─────────────────┘
│ HTTP
▼
┌─────────────────────────────────────────────────────────────────────────┐
│ lingma-ipc-proxy │
│ ┌─────────────────────────────────────────────────────────────────┐ │
│ │ internal/httpapi │ │
│ │ ┌─────────────┐ ┌─────────────┐ ┌─────────────────────────┐ │ │
│ │ │ /v1/models │ │/v1/chat/comp│ │ /v1/messages │ │ │
│ │ │ (GET) │ │ (POST) │ │ (POST) │ │ │
│ │ └──────┬──────┘ └──────┬──────┘ └───────────┬─────────────┘ │ │
│ │ └─────────────────┴──────────┬──────────┘ │ │
│ │ │ normalizeRequest │ │
│ │ ▼ │ │
│ │ ┌─────────────────────────────────────────────────────────┐ │ │
│ │ │ internal/service │ │ │
│ │ │ ┌──────────┐ ┌──────────┐ ┌────────────────────────┐ │ │ │
│ │ │ │ Session │ │ Prompt │ │ Stream/Event │ │ │ │
│ │ │ │ Manager │ │ Builder │ │ Handler │ │ │ │
│ │ │ └────┬─────┘ └────┬─────┘ └───────────┬────────────┘ │ │ │
│ │ │ └─────────────┴──────────┬─────────┘ │ │ │
│ │ │ │ buildLingmaPrompt │ │ │
│ │ │ ▼ │ │ │
│ │ │ ┌─────────────────────────────────────────────────┐ │ │ │
│ │ │ │ internal/lingmaipc │ │ │ │
│ │ │ │ ┌──────────────┐ ┌──────────────────────────┐ │ │ │ │
│ │ │ │ │ WebSocket │ │ Named Pipe (Win) │ │ │ │ │
│ │ │ │ │ Transport │ │ Transport │ │ │ │ │
│ │ │ │ └──────┬───────┘ └───────────┬──────────────┘ │ │ │ │
│ │ │ └─────────┼──────────────────────┼────────────────┘ │ │ │
│ │ └────────────┼──────────────────────┼────────────────────┘ │ │
│ │ │ │ │ │
│ │ ┌────────────┼──────────────────────┼────────────────────┐ │ │
│ │ │ ▼ ▼ │ │ │
│ │ │ ┌─────────────────────────────────────────────────┐ │ │ │
│ │ │ │ internal/toolemulation │ │ │ │
│ │ │ │ ┌──────────────┐ ┌──────────────────────────┐ │ │ │ │
│ │ │ │ │InjectTooling │ │ ParseActionBlocks │ │ │ │ │
│ │ │ │ │ (Prompt) │ │ (Response) │ │ │ │ │
│ │ │ │ └──────────────┘ └──────────────────────────┘ │ │ │ │
│ │ │ └─────────────────────────────────────────────────┘ │ │ │
│ │ └───────────────────────────────────────────────────────┘ │ │
│ └───────────────────────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────────────────────────┘
│
│ WebSocket / Named Pipe
▼
┌─────────────────────────────────────────────────────────────────────────┐
│ Lingma 后端进程 │
│ (VS Code 插件的本地 IPC 服务) │
│ ws://127.0.0.1:8899/ws │
└─────────────────────────────────────────────────────────────────────────┘
│
│ HTTP API
▼
┌─────────────────────────────────────────────────────────────────────────┐
│ 云端模型服务 │
│ (Kimi-K2.6 / Qwen3-Max / MiniMax-M2.7 等) │
└─────────────────────────────────────────────────────────────────────────┘
```mermaid
flowchart LR
A["客户端<br/>Claude Code / Hermes / Cline / Continue / OpenAI SDK / Anthropic SDK"]
B["internal/httpapi<br/>OpenAI + Anthropic 兼容路由"]
C["internal/service<br/>请求归一化 / 会话策略 / 流式输出 / 兜底"]
D["internal/toolemulation<br/>工具提示词注入 + action block 解析"]
E["internal/lingmaipc<br/>WebSocket / 命名管道"]
F["internal/remote<br/>登录态探测 / 模型列表 / Chat / SSE"]
G["Lingma 插件本地进程"]
H["Lingma 远端 API"]
I["桌面端 GUI<br/>Wails / 日志 / Token 统计 / 持久化状态"]
A --> B
I --> B
B --> C
C --> D
C --> E
C --> F
E --> G
F --> H
```
---
## 2. 模块职责
## 2. 运行模式
### 2.1 internal/httpapi
### 2.1 IPC 模式
HTTP API 适配层，负责将外部请求转换为内部 `service.ChatRequest`。
`backend=ipc`
| 端点 | 协议 | 功能 |
|------|------|------|
| `GET /v1/models` | OpenAI | 返回可用模型列表 |
| `POST /v1/chat/completions` | OpenAI | 聊天补全（流式/非流式） |
| `POST /v1/messages` | Anthropic | 消息接口（流式/非流式） |
- 读取本地 Lingma 插件传输信息
- 通过以下方式连接：
- macOS / Linux：WebSocket
- Windows：Named Pipe
- 复用 Lingma 插件自身的 session 语义
- 桌面端里“会话与环境”相关配置只在这里生效
**核心函数：**
- `handleOpenAIChatCompletions()` - 处理 OpenAI 格式请求
- `handleAnthropicMessages()` - 处理 Anthropic 格式请求
- `normalizeOpenAIRequest()` / `normalizeAnthropicRequest()` - 归一化请求
### 2.2 Remote API 模式
**关键设计：**
- 支持 CORS 预检请求 (`OPTIONS`)
- 单请求并发控制 (`tryAcquire()` / `release()`)
- 流式响应通过 `http.Flusher` 实现 SSE
`backend=remote`
### 2.2 internal/service
业务逻辑层，负责会话管理和 Prompt 构建。
**核心结构：**
```go
type Service struct {
cfg Config
client *lingmaipc.Client
stickySessionID string
stickyModelID string
}
```
**核心函数：**
- `Generate()` - 非流式生成
- `GenerateStream()` - 流式生成（返回 `events` + `done` channel）
- `buildLingmaPrompt()` - 构建 Lingma 原生 Prompt
- `runPromptLocked()` - 发送 `session/prompt` RPC 并监听 `session/update` 通知
**会话模式：**
| 模式 | 行为 |
|------|------|
| `reuse` | 复用 sticky session，多轮对话保持上下文 |
| `fresh` | 每个请求新建临时 session，完成后删除 |
| `auto` | 单轮请求复用；带 system/history 的请求用 fresh |
### 2.3 internal/lingmaipc
IPC 通信层，负责与 Lingma 后端进程建立连接。
**传输方式：**
| 平台 | 默认传输 | 说明 |
|------|----------|------|
| Windows | Named Pipe | `\\.\pipe\lingma-*` |
| macOS/Linux | WebSocket | `ws://127.0.0.1:{port}/ws` |
**连接发现：**
- 读取 VS Code 插件缓存：`~/.config/Lingma/SharedClientCache/.info.json`
- 获取 WebSocket 端口号
- 自动重连机制
**RPC 协议：**
- `session/new` - 创建会话
- `session/prompt` - 发送用户消息
- `session/update` - 接收流式响应通知
- `session/set_model` - 切换模型
- `chat/deleteSessionById` - 删除会话
### 2.4 internal/toolemulation
Tool 调用模拟层，将标准 `tools` 协议转换为 Prompt 层契约。
**核心流程：**
```
Client tools ──→ ExtractAnthropicTools() ──→ []Tool
│
▼
InjectTooling() ──→ System Prompt + Tool 说明
│
▼
模型输出 action block
│
▼
ParseActionBlocks() ──→ []ToolCall
│
▼
编码为 Anthropic tool_use / OpenAI tool_calls
```
**Prompt 契约格式：**
```
```json action
{"tool":"NAME","parameters":{"key":"value"}}
```
```
**支持格式：**
- `{"tool":"X","parameters":{}}` ✅ 标准格式
- `{"tool":"X","arguments":{}}` ✅ 兼容格式
- `{"tool":"X","input":{}}` ✅ 兼容格式
- `{"tool":"X","arg1":"val"}` ✅ 顶层参数（部分模型）
- 解析远端域名
- 加载认证信息：
- 显式指定的 `remote_auth_file`
- 或自动探测 `~/.lingma` 下的缓存
- 直接请求远端模型列表和聊天接口
- 支持远端超时 / 429 / 5xx 的模型兜底切换
- 不依赖本地插件会话环境参数
---
## 3. 核心流程
## 3. 模块职责
### 3.1 普通聊天请求流程
### 3.1 `cmd/lingma-ipc-proxy`
入口与配置装配层。
职责：
- 解析命令行参数
- 合并配置文件 / 环境变量 / CLI flags
- 选择后端模式
- 构建 `service.Config`
- 启动 `internal/httpapi.Server`
关键配置字段：
- `backend`
- `transport`
- `websocket_url`
- `pipe`
- `remote_base_url`
- `remote_auth_file`
- `remote_version`
- `remote_fallback_enabled`
- `remote_fallback_models`
### 3.2 `internal/httpapi`
OpenAI / Anthropic 兼容层。
主要路由：
- `GET /v1/models`
- `POST /v1/chat/completions`
- `POST /v1/messages`
- `GET /health`
- `GET /props`
职责：
- 把 OpenAI / Anthropic 请求归一化为 `service.ChatRequest`
- 把 service 结果重新编码成 OpenAI / Anthropic 响应
- 输出 SSE 流
- 记录调试用请求 / 响应摘要
### 3.3 `internal/service`
核心编排层。
职责：
- 选择当前 backend
- backend 预热
- 拉取模型列表
- 非流式生成
- 流式生成
- IPC 模式下的 session 复用策略
- 工具模拟注入与解析
- 图片输入归一化
- 远端 fallback 顺序控制
分支逻辑：
- IPC 路径走 `internal/lingmaipc`
- Remote 路径走 `internal/remote`
### 3.4 `internal/lingmaipc`
本地 Lingma 插件 IPC 客户端。
职责：
- 自动探测 WebSocket / pipe 端点
- 建立连接与重连
- 发送 RPC：
- `session/new`
- `session/prompt`
- `session/set_model`
- `chat/deleteSessionById`
- 消费 `session/update` 通知
### 3.5 `internal/remote`
Lingma 远端 HTTP 客户端。
职责：
- 解析远端 base URL
- 加载并校验登录态
- 生成远端请求所需身份信息
- 获取远端模型列表
- 调用远端聊天接口
- 处理远端 SSE 流式响应
### 3.6 `internal/toolemulation`
工具调用模拟层。
职责：
- 从 OpenAI / Anthropic 请求中提取工具定义
- 将工具契约注入 prompt
- 从模型文本里解析 JSON action block
- 回投为：
- Anthropic `tool_use`
- OpenAI `tool_calls`
---
## 4. 请求主流程
### 4.1 通用入口
```mermaid
sequenceDiagram
participant C as Client
participant H as HTTP API
participant S as Service
participant L as Lingma IPC
participant B as Lingma Backend
participant Client as Client
participant HTTP as httpapi
participant Service as service
C->>H: POST /v1/messages
H->>H: normalizeAnthropicRequest()
H->>S: GenerateStream(req)
S->>S: ensureConnected()
S->>S: resolveSession()
S->>S: buildLingmaPrompt()
S->>L: Send("session/prompt", params)
L->>B: WebSocket RPC
B->>L: session/update (agent_message_chunk)
loop 流式响应
L->>S: notification (chunk)
S->>H: events <- StreamEvent{Delta}
H->>C: SSE: content_block_delta
end
B->>L: session/update (chat_finish)
L->>S: notification (finish)
S->>H: done <- StreamResult
H->>C: SSE: message_stop
Client->>HTTP: OpenAI / Anthropic 请求
HTTP->>HTTP: 归一化请求
HTTP->>Service: Generate / GenerateStream
```
### 3.2 Tool 调用流程
### 4.2 IPC 后端流程
```mermaid
sequenceDiagram
participant C as Client
participant H as HTTP API
participant T as ToolEmulation
participant S as Service
participant L as Lingma IPC
participant Service as service
participant Tool as toolemulation
participant IPC as lingmaipc
participant Plugin as Lingma 插件
C->>H: POST /v1/messages (with tools)
H->>T: ExtractAnthropicTools()
H->>S: GenerateStream(req)
S->>T: InjectTooling(system, tools)
S->>L: session/prompt (with tool prompt)
L->>S: response (with action blocks)
S->>T: ParseActionBlocks(text)
T->>S: []ToolCall
S->>H: ChatResult{Text, ToolCalls}
H->>C: SSE: tool_use blocks
C->>H: POST /v1/messages (tool_result)
H->>T: ActionOutputPrompt(toolUseID, content)
H->>S: GenerateStream(req)
S->>L: session/prompt (with tool result)
L->>S: response
S->>H: ChatResult
H->>C: SSE: final response
Service->>Tool: 按需注入工具契约
Service->>IPC: ensure connected
Service->>IPC: 创建/复用 session
Service->>IPC: session/prompt
IPC->>Plugin: RPC
Plugin-->>IPC: session/update chunk
IPC-->>Service: 流式事件
Service-->>Service: 解析工具 block / 图片 / stop reason
```
### 3.3 图片传输流程
### 4.3 Remote 后端流程
```mermaid
sequenceDiagram
participant C as Client
participant H as HTTP API
participant S as Service
participant L as Lingma IPC
participant Service as service
participant Remote as remote client
participant API as Lingma 远端 API
C->>H: POST /v1/messages (with image)
H->>H: extractAnthropicImages()
H->>S: ChatRequest{Images: [...]}
S->>S: runPromptLocked()
Note over S: 1. 保存 base64 到 /tmp/lingma-img-*.ext
Note over S: 2. 构建 URI: lingma:///agent/file?path=...
S->>L: session/prompt
Note over L: prompt: [{type:"text"}, {type:"image", mimeType, uri, data}]
L->>S: response (model sees image)
S->>H: ChatResult
H->>C: SSE response
```
### 3.4 流式输出 SSE 事件序列
**Anthropic 格式（流式）：**
```
event: message_start
data: {"type":"message_start","message":{...}}
event: content_block_start
data: {"type":"content_block_start","index":0,"content_block":{"type":"text","text":""}}
event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"你"}}
event: content_block_delta
data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"好"}}
... (更多 delta)
event: content_block_stop
data: {"type":"content_block_stop","index":0}
[如有 tool_calls]
event: content_block_start
data: {"type":"content_block_start","index":1,"content_block":{"type":"tool_use","id":"...","name":"Bash","input":{"command":"ls /"}}}
event: content_block_stop
data: {"type":"content_block_stop","index":1}
event: message_delta
data: {"type":"message_delta","delta":{"stop_reason":"end_turn"},"usage":{"output_tokens":5}}
event: message_stop
data: {"type":"message_stop"}
Service->>Remote: 加载登录态 / 初始化 client
Service->>Remote: 需要时拉取模型列表
Service->>Remote: 发送 chat 请求
Remote->>API: HTTPS
API-->>Remote: JSON 或 SSE
Remote-->>Service: 归一化结果
Service-->>Service: 按条件执行 fallback
```
---
## 4. 关键技术决策
## 5. 远端兜底策略
### 4.1 为什么使用 Tool Emulation 而非原生 Tool Calling？
仅在以下条件同时满足时启用：
Lingma 后端模型（Kimi、Qwen 等）不原生支持 OpenAI/Anthropic 的 `tools` 协议。因此代理层需要将工具定义注入到 Prompt 中，通过结构化文本输出模拟工具调用。
- `backend=remote`
- `remote_fallback_enabled=true`
- 还没有向客户端输出任何流式 token
- 上游错误属于 timeout / 429 / 5xx
**优点：**
- 不依赖上游模型能力
- 兼容任何纯聊天模型
- 可精确控制 Prompt 格式
当前默认顺序：
**缺点：**
- 模型需要学习特定格式
- 解析可能有容错问题
- 增加了 Prompt 长度
1. `kmodel`
2. `mmodel`
3. `dashscope_qwen3_coder`
4. `dashscope_qmodel`
5. `dashscope_qwen_max_latest`
6. `dashscope_qwen_plus_20250428_thinking`
### 4.2 为什么使用 WebSocket/Named Pipe 而非 HTTP？
Lingma 插件使用本地 IPC 与后端通信，优势：
- 低延迟（本地通信）
- 双向实时通知（session/update）
- 认证信息由插件管理，代理无需处理
### 4.3 图片传输的双保险策略
```
Prompt 数组 (Lingma 原生格式):
[
{"type":"text","text":"..."},
{"type":"image","mimeType":"image/png","uri":"lingma:///agent/file?path=...","data":"base64..."}
]
```
- `uri`: Lingma 后端必须验证的本地文件路径
- `data`: base64 编码的图像数据（备用）
- `mimeType`: 图像类型标识
### 4.4 单请求并发控制
Lingma IPC 一次只能处理一个请求，因此代理使用 `tryAcquire()` 机制：
```go
if !s.tryAcquire() {
writeAnthropicError(w, 429, "rate_limit_error",
"Lingma IPC proxy handles one request at a time.")
return
}
defer s.release()
```
实际执行前，service 会先拿远端 `/v1/models` 的真实结果过滤一遍，只保留当前账号真的可用的模型。
---
## 5. 配置说明
## 6. 桌面端架构
### 5.1 配置文件结构
Wails 桌面端不是简单预览壳，而是本地代理的运维控制台。
```json
{
"host": "127.0.0.1",
"port": 8095,
"transport": "websocket",
"mode": "agent",
"shell_type": "zsh",
"session_mode": "auto",
"timeout": 120,
"cwd": "/Users/tiancheng"
}
```
职责：
### 5.2 配置项说明
- 启动 / 停止 / 重启代理
- 展示当前 backend、监听地址、探测结果
- 持久化：
- 请求历史
- 日志
- Token 统计
- 编辑配置并保存后按需重启
| 配置项 | 类型 | 默认值 | 说明 |
|--------|------|--------|------|
| `host` | string | `127.0.0.1` | HTTP 监听地址 |
| `port` | int | `8095` | HTTP 监听端口 |
| `transport` | string | `auto` | IPC 传输方式：`auto`/`pipe`/`websocket` |
| `mode` | string | `chat` | 模式：`chat`/`agent` |
| `shell_type` | string | `powershell` | 终端类型 |
| `session_mode` | string | `auto` | 会话模式：`reuse`/`fresh`/`auto` |
| `timeout` | int | `120` | 请求超时（秒） |
| `cwd` | string | `""` | 工作目录（传给 Lingma 后端） |
本地持久化路径：
- 配置：`~/.config/lingma-ipc-proxy/config.json`
- GUI 运行状态：`~/.config/lingma-ipc-proxy/app-state.json`
打包要求：
- 生产包不自动打开 Inspector / 调试入口
- 本地开发可通过 `LINGMA_DESKTOP_DEBUG=1` 显式开启
---
## 6. 扩展点
## 7. 关键设计决策
### 6.1 添加新模型
### 7.1 为什么同时保留 IPC 和 Remote？
在 `service.go` 的模型映射中添加：
因为两种模式解决的问题不同：
```go
func (s *Service) resolveInternalModelID(model string) string {
switch strings.ToLower(strings.TrimSpace(model)) {
case "kimi-k2.6":
return "kimi2.6"
case "qwen3-max":
return "qwen3max"
// 添加新模型映射
default:
return ""
}
}
```
- IPC 模式更贴近插件本地上下文和 session 语义
- Remote 模式更适合第三方 agent 客户端，减少对插件运行态的依赖
### 6.2 添加新 Tool 格式支持
### 7.2 为什么 Remote 也保留 Tool Emulation？
在 `toolemulation.go` 的 `parseToolCallJSON()` 中扩展参数解析逻辑。
因为 Lingma 暴露出来的模型能力并不保证始终稳定兼容 OpenAI / Anthropic 原生 tools 协议。代理层必须对外提供稳定契约，不能把上游模型差异直接泄露给客户端。
### 6.3 添加新 API 端点
### 7.3 为什么桌面端要持久化请求和 Token？
在 `httpapi/server.go` 的 `NewServer()` 中注册新路由。
因为这个 GUI 已经是运维面板，不是一次性调试页。重启后仍然需要保留最近请求、日志和 usage 统计，便于排障和观察模型表现。
---
*文档版本: 2025-04-25*
*对应代码版本: 当前 master*
## 8. 当前边界
- IPC 模式仍然受本地 Lingma 插件运行态影响
- Remote 登录态探测依赖本地 Lingma 缓存结构
- 图片类请求在本地持久化时会做裁剪/脱敏，避免状态文件过大
- Remote 模式下如果启用了 fallback，最近一次“聊天模型”可能与客户端最初指定模型不同
---
## 9. 代码入口建议
如果要继续扩展，优先看这些文件：
- `cmd/lingma-ipc-proxy/main.go`
- `internal/httpapi/server.go`
- `internal/service/service.go`
- `internal/lingmaipc/*`
- `internal/remote/*`
- `desktop/app.go`
- `desktop/main.go`
---
文档版本：2026-04-30

7
internal/httpapi/server.go
View File

@@ -1400,9 +1400,6 @@ func redactRecordedValue(value any) any {
if looksLikeImagePayload(typed) {
return imageRedaction(typed)
}
if len(typed) > 12000 {
return typed[:12000] + "... [truncated]"
}
return typed
default:
return typed
@@ -1443,12 +1440,8 @@ func mustMarshalJSON(value any) []byte {
}
func truncateRecordedString(value string) string {
const maxRecordedBody = 120000
if len(value) <= maxRecordedBody {
return value
}
return value[:maxRecordedBody] + "... [truncated]"
}
func withCORS(next http.Handler) http.Handler {
return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {

2
internal/remote/client.go
View File

@@ -76,7 +76,7 @@ func New(cfg Config) *Client {
cfg.CosyVersion = "2.11.2"
}
if cfg.Timeout <= 0 {
cfg.Timeout = 120 * time.Second
cfg.Timeout = 300 * time.Second
}
cfg.BaseURL = strings.TrimRight(cfg.BaseURL, "/")
return &Client{cfg: cfg, client: &http.Client{Timeout: cfg.Timeout}}

146
internal/service/service.go
View File

@@ -51,6 +51,8 @@ type Config struct {
ShellType string
SessionMode SessionMode
Timeout time.Duration
RemoteFallbackEnabled bool
RemoteFallbackModels []string
}
type Image struct {
@@ -166,7 +168,7 @@ func New(cfg Config) *Service {
cfg.ShellType = lingmaipc.DefaultShellType()
}
if cfg.Timeout <= 0 {
cfg.Timeout = 120 * time.Second
cfg.Timeout = 300 * time.Second
}
if cfg.Transport == "" {
cfg.Transport = lingmaipc.TransportAuto
@@ -174,6 +176,11 @@ func New(cfg Config) *Service {
if cfg.Backend == "" {
cfg.Backend = BackendRemote
}
if cfg.Backend == BackendRemote {
if len(cfg.RemoteFallbackModels) == 0 {
cfg.RemoteFallbackModels = DefaultRemoteFallbackModels()
}
}
cfg.Model = normalizeModelForBackend(cfg.Backend, cfg.Model)
if cfg.SessionMode == "" {
cfg.SessionMode = SessionModeAuto
@@ -181,6 +188,17 @@ func New(cfg Config) *Service {
return &Service{cfg: cfg}
}
func DefaultRemoteFallbackModels() []string {
return []string{
"kmodel",
"mmodel",
"dashscope_qwen3_coder",
"dashscope_qmodel",
"dashscope_qwen_max_latest",
"dashscope_qwen_plus_20250428_thinking",
}
}
func (s *Service) SetDefaultModel(model string) {
s.mu.Lock()
defer s.mu.Unlock()
@@ -331,9 +349,6 @@ func (s *Service) generateRemote(
req ChatRequest,
onDelta func(string),
) (*ChatResult, error) {
requestCtx, cancel := context.WithTimeout(ctx, s.cfg.Timeout)
defer cancel()
if strings.TrimSpace(req.Model) == "" {
req.Model = s.DefaultModel()
}
@@ -346,20 +361,54 @@ func (s *Service) generateRemote(
return nil, errors.New("empty user message")
}
models := s.remoteAttemptModels(ctx, req.Model)
client := s.remoteClientLocked()
remoteResult, err := client.Chat(requestCtx, remote.ChatRequest{
Model: req.Model,
var lastErr error
for i, model := range models {
attemptCtx, cancel := context.WithTimeout(ctx, s.cfg.Timeout)
result, emitted, err := s.generateRemoteWithModel(attemptCtx, client, req, prompt, model, onDelta)
cancel()
if err == nil {
return result, nil
}
lastErr = err
if i == len(models)-1 || emitted || !isRemoteFallbackError(err) {
return nil, err
}
}
return nil, lastErr
}
func (s *Service) generateRemoteWithModel(
ctx context.Context,
client *remote.Client,
req ChatRequest,
prompt string,
model string,
onDelta func(string),
) (*ChatResult, bool, error) {
emitted := false
delta := func(text string) {
if text != "" {
emitted = true
}
if onDelta != nil {
onDelta(text)
}
}
remoteResult, err := client.Chat(ctx, remote.ChatRequest{
Model: model,
Prompt: prompt,
Stream: onDelta != nil,
Temperature: req.Temperature,
}, onDelta)
}, delta)
if err != nil {
return nil, err
return nil, emitted, err
}
result := &ChatResult{
Text: remoteResult.Text,
Model: valueOr(strings.TrimSpace(req.Model), "lingma"),
Model: valueOr(strings.TrimSpace(model), "lingma"),
InputTokens: remoteResult.InputTokens,
OutputTokens: remoteResult.OutputTokens,
SessionID: "",
@@ -370,9 +419,9 @@ func (s *Service) generateRemote(
Transport: "remote",
EffectiveSession: SessionModeFresh,
}
s.applyToolEmulation(requestCtx, req, prompt, result, onDelta, func(hintPrompt string) (string, int, error) {
retryResult, retryErr := client.Chat(requestCtx, remote.ChatRequest{
Model: req.Model,
s.applyToolEmulation(ctx, req, prompt, result, onDelta, func(hintPrompt string) (string, int, error) {
retryResult, retryErr := client.Chat(ctx, remote.ChatRequest{
Model: model,
Prompt: hintPrompt,
Stream: onDelta != nil,
Temperature: req.Temperature,
@@ -385,7 +434,78 @@ func (s *Service) generateRemote(
}
return retryResult.Text, retryResult.OutputTokens, nil
})
return result, nil
return result, emitted, nil
}
func (s *Service) remoteAttemptModels(ctx context.Context, primary string) []string {
primary = normalizeModelForBackend(BackendRemote, primary)
models := []string{primary}
if !s.cfg.RemoteFallbackEnabled {
return models
}
availableCtx, cancel := context.WithTimeout(ctx, 15*time.Second)
remoteModels, err := s.remoteClientLocked().ListModels(availableCtx)
cancel()
if err != nil {
return models
}
available := make(map[string]bool, len(remoteModels))
for _, model := range remoteModels {
key := normalizeModelForBackend(BackendRemote, model.Key)
if key != "" {
available[key] = true
}
}
fallbackModels := s.cfg.RemoteFallbackModels
if len(fallbackModels) == 0 {
fallbackModels = DefaultRemoteFallbackModels()
}
ordered := make([]string, 0, len(fallbackModels))
seen := map[string]bool{primary: true}
primaryIndex := -1
for _, candidate := range fallbackModels {
model := normalizeModelForBackend(BackendRemote, candidate)
if model == "" {
continue
}
if model == primary && primaryIndex == -1 {
primaryIndex = len(ordered)
}
ordered = append(ordered, model)
}
start := 0
if primaryIndex >= 0 {
start = primaryIndex + 1
}
for _, model := range ordered[start:] {
if seen[model] || !available[model] {
continue
}
seen[model] = true
models = append(models, model)
}
return models
}
func isRemoteFallbackError(err error) bool {
if err == nil {
return false
}
if errors.Is(err, context.DeadlineExceeded) {
return true
}
msg := strings.ToLower(err.Error())
return strings.Contains(msg, "context deadline exceeded") ||
strings.Contains(msg, "client.timeout") ||
strings.Contains(msg, "timeout awaiting response") ||
strings.Contains(msg, "remote chat status 5") ||
strings.Contains(msg, "remote chat status 429") ||
strings.Contains(msg, "connection reset") ||
strings.Contains(msg, "unexpected eof")
}
func (s *Service) generateLocked(

11
lingma-ipc-proxy.macos.json
View File

@@ -7,6 +7,15 @@
"model": "kmodel",
"shell_type": "zsh",
"session_mode": "auto",
"timeout": 120,
"timeout": 300,
"remote_fallback_enabled": true,
"remote_fallback_models": [
"kmodel",
"mmodel",
"dashscope_qwen3_coder",
"dashscope_qmodel",
"dashscope_qwen_max_latest",
"dashscope_qwen_plus_20250428_thinking"
],
"cwd": "/Users/tiancheng"
}