@relayplane/proxy 0.1.7 → 0.1.9

package/README.md

@@ -176,13 +176,134 @@ Options:
   -h, --help         Show help
 ```
 
+## REST API
+
+The proxy exposes endpoints for stats and monitoring:
+
+### `GET /health`
+
+Server health and version info.
+
+```bash
+curl http://localhost:3001/health
+```
+
+```json
+{
+  "status": "ok",
+  "version": "0.1.7",
+  "uptime": "2h 15m 30s",
+  "providers": { "anthropic": true, "openai": true, "google": false },
+  "totalRuns": 142
+}
+```
+
+### `GET /stats`
+
+Aggregated statistics and cost savings.
+
+```bash
+curl http://localhost:3001/stats
+```
+
+```json
+{
+  "totalRuns": 142,
+  "savings": {
+    "estimatedSavingsPercent": "73.2%",
+    "actualCostUsd": "0.0234",
+    "baselineCostUsd": "0.0873",
+    "savedUsd": "0.0639"
+  },
+  "modelDistribution": {
+    "anthropic/claude-3-5-haiku-latest": { "count": 98, "percentage": "69.0%" },
+    "anthropic/claude-sonnet-4-20250514": { "count": 44, "percentage": "31.0%" }
+  }
+}
+```
+
+### `GET /runs`
+
+Recent routing decisions.
+
+```bash
+curl "http://localhost:3001/runs?limit=10"
+```
+
+```json
+{
+  "runs": [
+    {
+      "runId": "abc123",
+      "timestamp": "2026-02-03T13:26:03Z",
+      "model": "anthropic/claude-3-5-haiku-latest",
+      "taskType": "code_generation",
+      "confidence": 0.92,
+      "mode": "auto",
+      "durationMs": 1203,
+      "promptPreview": "Write a function that..."
+    }
+  ],
+  "total": 142
+}
+```
+
+## Configuration
+
+RelayPlane creates a config file on first run at `~/.relayplane/config.json`:
+
+```json
+{
+  "strategies": {
+    "code_review": { "model": "anthropic:claude-sonnet-4-20250514" },
+    "code_generation": { "model": "anthropic:claude-3-5-haiku-latest" },
+    "analysis": { "model": "anthropic:claude-sonnet-4-20250514" },
+    "summarization": { "model": "anthropic:claude-3-5-haiku-latest" },
+    "creative_writing": { "model": "anthropic:claude-sonnet-4-20250514" },
+    "data_extraction": { "model": "anthropic:claude-3-5-haiku-latest" },
+    "translation": { "model": "anthropic:claude-3-5-haiku-latest" },
+    "question_answering": { "model": "anthropic:claude-3-5-haiku-latest" },
+    "general": { "model": "anthropic:claude-3-5-haiku-latest" }
+  },
+  "defaults": {
+    "qualityModel": "claude-sonnet-4-20250514",
+    "costModel": "claude-3-5-haiku-latest"
+  }
+}
+```
+
+**Edit and save — changes apply instantly** (hot-reload, no restart needed).
+
+### Strategy Options
+
+| Field | Description |
+|-------|-------------|
+| `model` | Provider and model in format `provider:model` |
+| `minConfidence` | Optional. Only use this strategy if confidence >= threshold |
+| `fallback` | Optional. Fallback model if primary fails |
+
+### Examples
+
+Route all analysis tasks to GPT-4o:
+```json
+"analysis": { "model": "openai:gpt-4o" }
+```
+
+Use Opus for code review with fallback:
+```json
+"code_review": { 
+  "model": "anthropic:claude-opus-4-5-20250514",
+  "fallback": "anthropic:claude-sonnet-4-20250514"
+}
+```
+
 ## Data Storage
 
 All data stored locally at `~/.relayplane/data.db` (SQLite).
 
 ```bash
 # View recent runs
-sqlite3 ~/.relayplane/data.db "SELECT * FROM runs ORDER BY timestamp DESC LIMIT 10"
+sqlite3 ~/.relayplane/data.db "SELECT * FROM runs ORDER BY created_at DESC LIMIT 10"
 
 # Check routing rules
 sqlite3 ~/.relayplane/data.db "SELECT * FROM routing_rules"

package/dist/cli.js

@@ -1591,12 +1591,127 @@ ${input.prompt}` : input.prompt;
   }
 };
 
+// src/config.ts
+var fs2 = __toESM(require("fs"));
+var path2 = __toESM(require("path"));
+var os2 = __toESM(require("os"));
+var import_zod = require("zod");
+var StrategySchema = import_zod.z.object({
+  model: import_zod.z.string(),
+  minConfidence: import_zod.z.number().min(0).max(1).optional(),
+  fallback: import_zod.z.string().optional()
+});
+var AuthSchema = import_zod.z.object({
+  anthropicApiKey: import_zod.z.string().optional(),
+  anthropicMaxToken: import_zod.z.string().optional(),
+  useMaxForModels: import_zod.z.array(import_zod.z.string()).optional()
+  // Default: ['opus']
+}).optional();
+var ConfigSchema = import_zod.z.object({
+  strategies: import_zod.z.record(import_zod.z.string(), StrategySchema).optional(),
+  defaults: import_zod.z.object({
+    qualityModel: import_zod.z.string().optional(),
+    costModel: import_zod.z.string().optional()
+  }).optional(),
+  auth: AuthSchema
+});
+var DEFAULT_CONFIG = {
+  strategies: {
+    code_review: { model: "anthropic:claude-sonnet-4-20250514" },
+    code_generation: { model: "anthropic:claude-3-5-haiku-latest" },
+    analysis: { model: "anthropic:claude-sonnet-4-20250514" },
+    summarization: { model: "anthropic:claude-3-5-haiku-latest" },
+    creative_writing: { model: "anthropic:claude-sonnet-4-20250514" },
+    data_extraction: { model: "anthropic:claude-3-5-haiku-latest" },
+    translation: { model: "anthropic:claude-3-5-haiku-latest" },
+    question_answering: { model: "anthropic:claude-3-5-haiku-latest" },
+    general: { model: "anthropic:claude-3-5-haiku-latest" }
+  },
+  defaults: {
+    qualityModel: "claude-sonnet-4-20250514",
+    costModel: "claude-3-5-haiku-latest"
+  }
+};
+function getConfigPath() {
+  return path2.join(os2.homedir(), ".relayplane", "config.json");
+}
+function writeDefaultConfig() {
+  const configPath = getConfigPath();
+  const dir = path2.dirname(configPath);
+  if (!fs2.existsSync(dir)) {
+    fs2.mkdirSync(dir, { recursive: true });
+  }
+  if (!fs2.existsSync(configPath)) {
+    fs2.writeFileSync(
+      configPath,
+      JSON.stringify(DEFAULT_CONFIG, null, 2) + "\n",
+      "utf-8"
+    );
+    console.log(`[relayplane] Created default config at ${configPath}`);
+  }
+}
+function loadConfig() {
+  const configPath = getConfigPath();
+  writeDefaultConfig();
+  try {
+    const raw = fs2.readFileSync(configPath, "utf-8");
+    const parsed = JSON.parse(raw);
+    const validated = ConfigSchema.parse(parsed);
+    return validated;
+  } catch (err) {
+    if (err instanceof import_zod.z.ZodError) {
+      console.error(`[relayplane] Invalid config: ${err.message}`);
+    } else if (err instanceof SyntaxError) {
+      console.error(`[relayplane] Config JSON parse error: ${err.message}`);
+    } else {
+      console.error(`[relayplane] Failed to load config: ${err}`);
+    }
+    console.log("[relayplane] Using default config");
+    return DEFAULT_CONFIG;
+  }
+}
+function getStrategy(config, taskType) {
+  return config.strategies?.[taskType] ?? null;
+}
+function getAnthropicAuth(config, model) {
+  const auth = config.auth;
+  const useMaxForModels = auth?.useMaxForModels ?? ["opus"];
+  const shouldUseMax = useMaxForModels.some((m) => model.toLowerCase().includes(m.toLowerCase()));
+  if (shouldUseMax && auth?.anthropicMaxToken) {
+    return { type: "max", value: auth.anthropicMaxToken };
+  }
+  const apiKey = auth?.anthropicApiKey ?? process.env["ANTHROPIC_API_KEY"];
+  if (apiKey) {
+    return { type: "apiKey", value: apiKey };
+  }
+  return null;
+}
+function watchConfig(onChange) {
+  const configPath = getConfigPath();
+  const dir = path2.dirname(configPath);
+  if (!fs2.existsSync(dir)) {
+    fs2.mkdirSync(dir, { recursive: true });
+  }
+  let debounceTimer = null;
+  fs2.watch(dir, (eventType, filename) => {
+    if (filename === "config.json") {
+      if (debounceTimer) clearTimeout(debounceTimer);
+      debounceTimer = setTimeout(() => {
+        console.log("[relayplane] Config file changed, reloading...");
+        const newConfig = loadConfig();
+        onChange(newConfig);
+      }, 100);
+    }
+  });
+}
+
 // src/proxy.ts
-var VERSION = "0.1.7";
+var VERSION = "0.1.9";
 var recentRuns = [];
 var MAX_RECENT_RUNS = 100;
 var modelCounts = {};
 var serverStartTime = 0;
+var currentConfig = loadConfig();
 var DEFAULT_ENDPOINTS = {
   anthropic: {
     baseUrl: "https://api.anthropic.com/v1",
@@ -1659,13 +1774,17 @@ function extractPromptText(messages) {
     return "";
   }).join("\n");
 }
-async function forwardToAnthropic(request, targetModel, apiKey, betaHeaders) {
+async function forwardToAnthropic(request, targetModel, auth, betaHeaders) {
   const anthropicBody = buildAnthropicBody(request, targetModel, false);
   const headers = {
     "Content-Type": "application/json",
-    "x-api-key": apiKey,
     "anthropic-version": "2023-06-01"
   };
+  if (auth.type === "max") {
+    headers["Authorization"] = `Bearer ${auth.value}`;
+  } else {
+    headers["x-api-key"] = auth.value;
+  }
   if (betaHeaders) {
     headers["anthropic-beta"] = betaHeaders;
   }
@@ -1676,13 +1795,17 @@ async function forwardToAnthropic(request, targetModel, apiKey, betaHeaders) {
   });
   return response;
 }
-async function forwardToAnthropicStream(request, targetModel, apiKey, betaHeaders) {
+async function forwardToAnthropicStream(request, targetModel, auth, betaHeaders) {
   const anthropicBody = buildAnthropicBody(request, targetModel, true);
   const headers = {
     "Content-Type": "application/json",
-    "x-api-key": apiKey,
     "anthropic-version": "2023-06-01"
   };
+  if (auth.type === "max") {
+    headers["Authorization"] = `Bearer ${auth.value}`;
+  } else {
+    headers["x-api-key"] = auth.value;
+  }
   if (betaHeaders) {
     headers["anthropic-beta"] = betaHeaders;
   }
@@ -2447,42 +2570,65 @@ async function startProxy(config = {}) {
     const confidence = getInferenceConfidence(promptText, taskType);
     log(`Inferred task: ${taskType} (confidence: ${confidence.toFixed(2)})`);
     if (routingMode !== "passthrough") {
-      const rule = relay.routing.get(taskType);
-      if (rule && rule.preferredModel) {
-        const parsed = parsePreferredModel(rule.preferredModel);
+      const configStrategy = getStrategy(currentConfig, taskType);
+      if (configStrategy) {
+        const parsed = parsePreferredModel(configStrategy.model);
         if (parsed) {
           targetProvider = parsed.provider;
           targetModel = parsed.model;
-          log(`Using learned rule: ${rule.preferredModel}`);
+          log(`Using config strategy: ${configStrategy.model}`);
+        }
+      }
+      if (!configStrategy) {
+        const rule = relay.routing.get(taskType);
+        if (rule && rule.preferredModel) {
+          const parsed = parsePreferredModel(rule.preferredModel);
+          if (parsed) {
+            targetProvider = parsed.provider;
+            targetModel = parsed.model;
+            log(`Using learned rule: ${rule.preferredModel}`);
+          } else {
+            const defaultRoute = DEFAULT_ROUTING[taskType];
+            targetProvider = defaultRoute.provider;
+            targetModel = defaultRoute.model;
+          }
         } else {
           const defaultRoute = DEFAULT_ROUTING[taskType];
           targetProvider = defaultRoute.provider;
           targetModel = defaultRoute.model;
         }
-      } else {
-        const defaultRoute = DEFAULT_ROUTING[taskType];
-        targetProvider = defaultRoute.provider;
-        targetModel = defaultRoute.model;
       }
       if (routingMode === "cost") {
-        const simpleTasks = ["summarization", "data_extraction", "translation", "question_answering"];
-        if (simpleTasks.includes(taskType)) {
-          targetModel = "claude-3-5-haiku-latest";
-          targetProvider = "anthropic";
-        }
+        const costModel = currentConfig.defaults?.costModel || "claude-3-5-haiku-latest";
+        targetModel = costModel;
+        targetProvider = "anthropic";
+        log(`Cost mode: using ${costModel}`);
       } else if (routingMode === "quality") {
-        const qualityModel = process.env["RELAYPLANE_QUALITY_MODEL"] || "claude-sonnet-4-20250514";
+        const qualityModel = currentConfig.defaults?.qualityModel || process.env["RELAYPLANE_QUALITY_MODEL"] || "claude-sonnet-4-20250514";
         targetModel = qualityModel;
         targetProvider = "anthropic";
+        log(`Quality mode: using ${qualityModel}`);
       }
     }
     log(`Routing to: ${targetProvider}/${targetModel}`);
-    const apiKeyEnv = DEFAULT_ENDPOINTS[targetProvider]?.apiKeyEnv ?? `${targetProvider.toUpperCase()}_API_KEY`;
-    const apiKey = process.env[apiKeyEnv];
-    if (!apiKey) {
-      res.writeHead(500, { "Content-Type": "application/json" });
-      res.end(JSON.stringify({ error: `Missing ${apiKeyEnv} environment variable` }));
-      return;
+    let apiKey;
+    let anthropicAuth = null;
+    if (targetProvider === "anthropic") {
+      anthropicAuth = getAnthropicAuth(currentConfig, targetModel);
+      if (!anthropicAuth) {
+        res.writeHead(500, { "Content-Type": "application/json" });
+        res.end(JSON.stringify({ error: "No Anthropic auth configured (set ANTHROPIC_API_KEY or config.auth.anthropicMaxToken)" }));
+        return;
+      }
+      log(`Using ${anthropicAuth.type === "max" ? "MAX token" : "API key"} auth for ${targetModel}`);
+    } else {
+      const apiKeyEnv = DEFAULT_ENDPOINTS[targetProvider]?.apiKeyEnv ?? `${targetProvider.toUpperCase()}_API_KEY`;
+      apiKey = process.env[apiKeyEnv];
+      if (!apiKey) {
+        res.writeHead(500, { "Content-Type": "application/json" });
+        res.end(JSON.stringify({ error: `Missing ${apiKeyEnv} environment variable` }));
+        return;
+      }
     }
     const startTime = Date.now();
     const betaHeaders = req.headers["anthropic-beta"];
@@ -2493,6 +2639,7 @@ async function startProxy(config = {}) {
         targetProvider,
         targetModel,
         apiKey,
+        anthropicAuth,
         relay,
         promptText,
         taskType,
@@ -2509,6 +2656,7 @@ async function startProxy(config = {}) {
         targetProvider,
         targetModel,
         apiKey,
+        anthropicAuth,
         relay,
         promptText,
         taskType,
@@ -2520,6 +2668,10 @@ async function startProxy(config = {}) {
       );
     }
   });
+  watchConfig((newConfig) => {
+    currentConfig = newConfig;
+    console.log("[relayplane] Config reloaded");
+  });
   return new Promise((resolve, reject) => {
     server.on("error", reject);
     server.listen(port, host, () => {
@@ -2528,17 +2680,19 @@ async function startProxy(config = {}) {
       console.log(`  Models: relayplane:auto, relayplane:cost, relayplane:quality`);
       console.log(`  Endpoint: POST /v1/chat/completions`);
       console.log(`  Stats: GET /stats, /runs, /health`);
+      console.log(`  Config: ~/.relayplane/config.json (hot-reload enabled)`);
       console.log(`  Streaming: \u2705 Enabled`);
       resolve(server);
     });
   });
 }
-async function handleStreamingRequest(res, request, targetProvider, targetModel, apiKey, relay, promptText, taskType, confidence, routingMode, startTime, log, betaHeaders) {
+async function handleStreamingRequest(res, request, targetProvider, targetModel, apiKey, anthropicAuth, relay, promptText, taskType, confidence, routingMode, startTime, log, betaHeaders) {
   let providerResponse;
   try {
     switch (targetProvider) {
       case "anthropic":
-        providerResponse = await forwardToAnthropicStream(request, targetModel, apiKey, betaHeaders);
+        if (!anthropicAuth) throw new Error("No Anthropic auth");
+        providerResponse = await forwardToAnthropicStream(request, targetModel, anthropicAuth, betaHeaders);
         break;
       case "google":
         providerResponse = await forwardToGeminiStream(request, targetModel, apiKey);
@@ -2616,13 +2770,14 @@ async function handleStreamingRequest(res, request, targetProvider, targetModel,
   });
   res.end();
 }
-async function handleNonStreamingRequest(res, request, targetProvider, targetModel, apiKey, relay, promptText, taskType, confidence, routingMode, startTime, log, betaHeaders) {
+async function handleNonStreamingRequest(res, request, targetProvider, targetModel, apiKey, anthropicAuth, relay, promptText, taskType, confidence, routingMode, startTime, log, betaHeaders) {
   let providerResponse;
   let responseData;
   try {
     switch (targetProvider) {
       case "anthropic": {
-        providerResponse = await forwardToAnthropic(request, targetModel, apiKey, betaHeaders);
+        if (!anthropicAuth) throw new Error("No Anthropic auth");
+        providerResponse = await forwardToAnthropic(request, targetModel, anthropicAuth, betaHeaders);
         const rawData = await providerResponse.json();
         if (!providerResponse.ok) {
           res.writeHead(providerResponse.status, { "Content-Type": "application/json" });