@blockrun/clawrouter 0.9.14 → 0.9.16

package/dist/index.js

@@ -3479,6 +3479,174 @@ var PROXY_PORT = (() => {
   return DEFAULT_PORT;
 })();
 
+// src/journal.ts
+var DEFAULT_CONFIG2 = {
+  maxEntries: 100,
+  maxAgeMs: 24 * 60 * 60 * 1e3,
+  // 24 hours
+  maxEventsPerResponse: 5
+};
+var SessionJournal = class {
+  journals = /* @__PURE__ */ new Map();
+  config;
+  constructor(config) {
+    this.config = { ...DEFAULT_CONFIG2, ...config };
+  }
+  /**
+   * Extract key events from assistant response content.
+   * Looks for patterns like "I created...", "I fixed...", "Successfully..."
+   */
+  extractEvents(content) {
+    if (!content || typeof content !== "string") {
+      return [];
+    }
+    const events = [];
+    const seen = /* @__PURE__ */ new Set();
+    const patterns = [
+      // Creation patterns
+      /I (?:also |then |have |)?(?:created|implemented|added|wrote|built|generated|set up|initialized) ([^.!?\n]{10,150})/gi,
+      // Fix patterns
+      /I (?:also |then |have |)?(?:fixed|resolved|solved|patched|corrected|addressed|debugged) ([^.!?\n]{10,150})/gi,
+      // Completion patterns
+      /I (?:also |then |have |)?(?:completed|finished|done with|wrapped up) ([^.!?\n]{10,150})/gi,
+      // Update patterns
+      /I (?:also |then |have |)?(?:updated|modified|changed|refactored|improved|enhanced|optimized) ([^.!?\n]{10,150})/gi,
+      // Success patterns
+      /Successfully ([^.!?\n]{10,150})/gi,
+      // Tool usage patterns (when agent uses tools)
+      /I (?:also |then |have |)?(?:ran|executed|called|invoked) ([^.!?\n]{10,100})/gi
+    ];
+    for (const pattern of patterns) {
+      pattern.lastIndex = 0;
+      let match;
+      while ((match = pattern.exec(content)) !== null) {
+        const action = match[0].trim();
+        const normalized = action.toLowerCase();
+        if (seen.has(normalized)) {
+          continue;
+        }
+        if (action.length >= 15 && action.length <= 200) {
+          events.push(action);
+          seen.add(normalized);
+        }
+        if (events.length >= this.config.maxEventsPerResponse) {
+          break;
+        }
+      }
+      if (events.length >= this.config.maxEventsPerResponse) {
+        break;
+      }
+    }
+    return events;
+  }
+  /**
+   * Record events to the session journal.
+   */
+  record(sessionId, events, model) {
+    if (!sessionId || !events.length) {
+      return;
+    }
+    const journal = this.journals.get(sessionId) || [];
+    const now = Date.now();
+    for (const action of events) {
+      journal.push({
+        timestamp: now,
+        action,
+        model
+      });
+    }
+    const cutoff = now - this.config.maxAgeMs;
+    const trimmed = journal.filter((e) => e.timestamp > cutoff).slice(-this.config.maxEntries);
+    this.journals.set(sessionId, trimmed);
+  }
+  /**
+   * Check if the user message indicates a need for historical context.
+   */
+  needsContext(lastUserMessage) {
+    if (!lastUserMessage || typeof lastUserMessage !== "string") {
+      return false;
+    }
+    const lower = lastUserMessage.toLowerCase();
+    const triggers = [
+      // Direct questions about past work
+      "what did you do",
+      "what have you done",
+      "what did we do",
+      "what have we done",
+      // Temporal references
+      "earlier",
+      "before",
+      "previously",
+      "this session",
+      "today",
+      "so far",
+      // Summary requests
+      "remind me",
+      "summarize",
+      "summary of",
+      "recap",
+      // Progress inquiries
+      "your work",
+      "your progress",
+      "accomplished",
+      "achievements",
+      "completed tasks"
+    ];
+    return triggers.some((t) => lower.includes(t));
+  }
+  /**
+   * Format the journal for injection into system message.
+   * Returns null if journal is empty.
+   */
+  format(sessionId) {
+    const journal = this.journals.get(sessionId);
+    if (!journal?.length) {
+      return null;
+    }
+    const lines = journal.map((e) => {
+      const time = new Date(e.timestamp).toLocaleTimeString("en-US", {
+        hour: "2-digit",
+        minute: "2-digit",
+        hour12: true
+      });
+      return `- ${time}: ${e.action}`;
+    });
+    return `[Session Memory - Key Actions]
+${lines.join("\n")}`;
+  }
+  /**
+   * Get the raw journal entries for a session (for debugging/testing).
+   */
+  getEntries(sessionId) {
+    return this.journals.get(sessionId) || [];
+  }
+  /**
+   * Clear journal for a specific session.
+   */
+  clear(sessionId) {
+    this.journals.delete(sessionId);
+  }
+  /**
+   * Clear all journals.
+   */
+  clearAll() {
+    this.journals.clear();
+  }
+  /**
+   * Get stats about the journal.
+   */
+  getStats() {
+    let totalEntries = 0;
+    for (const entries of this.journals.values()) {
+      totalEntries += entries.length;
+    }
+    return {
+      sessions: this.journals.size,
+      totalEntries
+    };
+  }
+};
+
 // src/proxy.ts
 var BLOCKRUN_API = "https://blockrun.ai/api";
 var AUTO_MODEL = "blockrun/auto";
@@ -3888,6 +4056,7 @@ async function startProxy(options) {
   const deduplicator = new RequestDeduplicator();
   const responseCache = new ResponseCache(options.cacheConfig);
   const sessionStore = new SessionStore(options.sessionConfig);
+  const sessionJournal = new SessionJournal();
   const connections = /* @__PURE__ */ new Set();
   const server = createServer(async (req, res) => {
     req.on("error", (err) => {
@@ -3983,7 +4152,8 @@ async function startProxy(options) {
         deduplicator,
         balanceMonitor,
         sessionStore,
-        responseCache
+        responseCache,
+        sessionJournal
       );
     } catch (err) {
       const error = err instanceof Error ? err : new Error(String(err));
@@ -4184,7 +4354,7 @@ async function tryModelRequest(upstreamUrl, method, headers, body, modelId, maxT
     };
   }
 }
-async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, deduplicator, balanceMonitor, sessionStore, responseCache) {
+async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, deduplicator, balanceMonitor, sessionStore, responseCache, sessionJournal) {
   const startTime = Date.now();
   const upstreamUrl = `${apiBase}${req.url}`;
   const bodyChunks = [];
@@ -4197,7 +4367,9 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
   let modelId = "";
   let maxTokens = 4096;
   let routingProfile = null;
+  let accumulatedContent = "";
   const isChatCompletion = req.url?.includes("/chat/completions");
+  const sessionId = getSessionId(req.headers);
   if (isChatCompletion && body.length > 0) {
     try {
       const parsed = JSON.parse(body.toString());
@@ -4205,6 +4377,30 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
       modelId = parsed.model || "";
       maxTokens = parsed.max_tokens || 4096;
       let bodyModified = false;
+      if (sessionId && Array.isArray(parsed.messages)) {
+        const messages = parsed.messages;
+        const lastUserMsg = [...messages].reverse().find((m) => m.role === "user");
+        const lastContent = typeof lastUserMsg?.content === "string" ? lastUserMsg.content : "";
+        if (sessionJournal.needsContext(lastContent)) {
+          const journalText = sessionJournal.format(sessionId);
+          if (journalText) {
+            const sysIdx = messages.findIndex((m) => m.role === "system");
+            if (sysIdx >= 0 && typeof messages[sysIdx].content === "string") {
+              messages[sysIdx] = {
+                ...messages[sysIdx],
+                content: journalText + "\n\n" + messages[sysIdx].content
+              };
+            } else {
+              messages.unshift({ role: "system", content: journalText });
+            }
+            parsed.messages = messages;
+            bodyModified = true;
+            console.log(
+              `[ClawRouter] Injected session journal (${journalText.length} chars) for session ${sessionId.slice(0, 8)}...`
+            );
+          }
+        }
+      }
       if (parsed.stream === true) {
         parsed.stream = false;
         bodyModified = true;
@@ -4243,18 +4439,18 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
             latencyMs: 0
           });
         } else {
-          const sessionId = getSessionId(
+          const sessionId2 = getSessionId(
             req.headers
           );
-          const existingSession = sessionId ? sessionStore.getSession(sessionId) : void 0;
+          const existingSession = sessionId2 ? sessionStore.getSession(sessionId2) : void 0;
           if (existingSession) {
             console.log(
-              `[ClawRouter] Session ${sessionId?.slice(0, 8)}... using pinned model: ${existingSession.model}`
+              `[ClawRouter] Session ${sessionId2?.slice(0, 8)}... using pinned model: ${existingSession.model}`
             );
             parsed.model = existingSession.model;
             modelId = existingSession.model;
             bodyModified = true;
-            sessionStore.touchSession(sessionId);
+            sessionStore.touchSession(sessionId2);
           } else {
             const messages = parsed.messages;
             let lastUserMsg;
@@ -4283,10 +4479,10 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
             parsed.model = routingDecision.model;
             modelId = routingDecision.model;
             bodyModified = true;
-            if (sessionId) {
-              sessionStore.setSession(sessionId, routingDecision.model, routingDecision.tier);
+            if (sessionId2) {
+              sessionStore.setSession(sessionId2, routingDecision.model, routingDecision.tier);
               console.log(
-                `[ClawRouter] Session ${sessionId.slice(0, 8)}... pinned to model: ${routingDecision.model}`
+                `[ClawRouter] Session ${sessionId2.slice(0, 8)}... pinned to model: ${routingDecision.model}`
               );
             }
             options.onRouted?.(routingDecision);
@@ -4621,6 +4817,9 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
               const content = stripThinkingTokens(rawContent);
               const role = choice.message?.role ?? choice.delta?.role ?? "assistant";
               const index = choice.index ?? 0;
+              if (content) {
+                accumulatedContent += content;
+              }
               const roleChunk = {
                 ...baseChunk,
                 choices: [{ index, delta: { role }, logprobs: null, finish_reason: null }]
@@ -4734,6 +4933,22 @@ async function proxyRequest(req, res, apiBase, payFetch, options, routerOpts, de
         });
         console.log(`[ClawRouter] Cached response for ${modelId} (${responseBody.length} bytes)`);
       }
+      try {
+        const rspJson = JSON.parse(responseBody.toString());
+        if (rspJson.choices?.[0]?.message?.content) {
+          accumulatedContent = rspJson.choices[0].message.content;
+        }
+      } catch {
+      }
+    }
+    if (sessionId && accumulatedContent) {
+      const events = sessionJournal.extractEvents(accumulatedContent);
+      if (events.length > 0) {
+        sessionJournal.record(sessionId, events, actualModelUsed);
+        console.log(
+          `[ClawRouter] Recorded ${events.length} events to session journal for session ${sessionId.slice(0, 8)}...`
+        );
+      }
     }
     if (estimatedCostMicros !== void 0) {
       balanceMonitor.deductEstimated(estimatedCostMicros);
@@ -5044,7 +5259,7 @@ function injectModelsConfig(logger) {
     { id: "opus", alias: "br-opus" },
     { id: "haiku", alias: "br-haiku" },
     { id: "gpt5", alias: "gpt5" },
-    { id: "mini", alias: "mini" },
+    { id: "codex", alias: "codex" },
     { id: "grok-fast", alias: "grok-fast" },
     { id: "grok-code", alias: "grok-code" },
     { id: "deepseek", alias: "deepseek" },
@@ -5053,7 +5268,7 @@ function injectModelsConfig(logger) {
     { id: "gemini", alias: "gemini" },
     { id: "flash", alias: "flash" }
   ];
-  const DEPRECATED_ALIASES = ["blockrun/nvidia", "blockrun/gpt", "blockrun/o3", "blockrun/grok"];
+  const DEPRECATED_ALIASES = ["blockrun/nvidia", "blockrun/gpt", "blockrun/o3", "blockrun/grok", "blockrun/mini"];
   if (!defaults.models) {
     defaults.models = {};
     needsWrite = true;