kimiflare 0.34.0 → 0.34.1

package/dist/index.js

@@ -83,7 +83,7 @@ async function loadConfig() {
   const envCacheStable = process.env.KIMIFLARE_CACHE_STABLE_PROMPTS;
   const cacheStablePrompts = envCacheStable === "0" || envCacheStable === "false" ? false : true;
   const envCompiled = process.env.KIMIFLARE_COMPILED_CONTEXT;
-  const compiledContext = envCompiled === "1" || envCompiled === "true" ? true : false;
+  const compiledContext = envCompiled === "0" || envCompiled === "false" ? false : true;
   const envImageTurns = process.env.KIMIFLARE_IMAGE_HISTORY_TURNS;
   const imageHistoryTurns = envImageTurns ? parseInt(envImageTurns, 10) : void 0;
   const envMemoryEnabled = readBooleanEnv("KIMIFLARE_MEMORY_ENABLED");
@@ -92,6 +92,7 @@ async function loadConfig() {
   const envMemoryMaxEntries = readNumberEnv("KIMIFLARE_MEMORY_MAX_ENTRIES");
   const envMemoryEmbeddingModel = process.env.KIMIFLARE_MEMORY_EMBEDDING_MODEL;
   const envPlumbingModel = process.env.KIMIFLARE_PLUMBING_MODEL;
+  const envMemoryExtractionModel = process.env.KIMIFLARE_MEMORY_EXTRACTION_MODEL;
   const envCodeMode = readBooleanEnv("KIMIFLARE_CODE_MODE");
   const envCostAttribution = readBooleanEnv("KIMI_COST_ATTRIBUTION");
   const envFilePicker = readBooleanEnv("KIMIFLARE_FILE_PICKER");
@@ -112,14 +113,15 @@ async function loadConfig() {
       cacheStablePrompts,
       compiledContext,
       imageHistoryTurns: Number.isNaN(imageHistoryTurns) ? void 0 : imageHistoryTurns,
-      memoryEnabled: envMemoryEnabled,
+      memoryEnabled: envMemoryEnabled ?? true,
       memoryDbPath: envMemoryDbPath,
       memoryMaxAgeDays: envMemoryMaxAgeDays,
       memoryMaxEntries: envMemoryMaxEntries,
       memoryEmbeddingModel: envMemoryEmbeddingModel,
       plumbingModel: envPlumbingModel,
-      codeMode: envCodeMode,
-      costAttribution: envCostAttribution ?? false,
+      memoryExtractionModel: envMemoryExtractionModel,
+      codeMode: envCodeMode ?? true,
+      costAttribution: envCostAttribution ?? true,
       filePicker: envFilePicker ?? true
     };
   }
@@ -144,14 +146,15 @@ async function loadConfig() {
         cacheStablePrompts: parsed.cacheStablePrompts ?? cacheStablePrompts,
         compiledContext: parsed.compiledContext ?? compiledContext,
         imageHistoryTurns: Number.isNaN(imageHistoryTurns) ? parsed.imageHistoryTurns : imageHistoryTurns,
-        memoryEnabled: envMemoryEnabled ?? parsed.memoryEnabled,
+        memoryEnabled: envMemoryEnabled ?? parsed.memoryEnabled ?? true,
         memoryDbPath: envMemoryDbPath ?? parsed.memoryDbPath,
         memoryMaxAgeDays: envMemoryMaxAgeDays ?? parsed.memoryMaxAgeDays,
         memoryMaxEntries: envMemoryMaxEntries ?? parsed.memoryMaxEntries,
         memoryEmbeddingModel: envMemoryEmbeddingModel ?? parsed.memoryEmbeddingModel,
         plumbingModel: envPlumbingModel ?? parsed.plumbingModel,
-        codeMode: envCodeMode ?? parsed.codeMode,
-        costAttribution: envCostAttribution ?? parsed.costAttribution ?? false,
+        memoryExtractionModel: envMemoryExtractionModel ?? parsed.memoryExtractionModel,
+        codeMode: envCodeMode ?? parsed.codeMode ?? true,
+        costAttribution: envCostAttribution ?? parsed.costAttribution ?? true,
         filePicker: envFilePicker ?? parsed.filePicker ?? true
       };
     }
@@ -553,44 +556,51 @@ async function* parseStream(body, signal) {
       lastUsage = chunk.usage;
       yield { type: "usage", usage: chunk.usage };
     }
+    if (typeof chunk.response === "string") {
+      const resp = chunk.response;
+      if (resp.length) {
+        yield { type: "text", delta: resp };
+      }
+    }
     const choice = chunk.choices?.[0];
-    if (!choice) continue;
-    const d = choice.delta;
-    if (d) {
-      if (typeof d.reasoning_content === "string" && d.reasoning_content.length) {
-        yield { type: "reasoning", delta: d.reasoning_content };
-      }
-      if (typeof d.content === "string" && d.content.length) {
-        yield { type: "text", delta: d.content };
-      }
-      if (Array.isArray(d.tool_calls)) {
-        for (const tc of d.tool_calls) {
-          const idx = typeof tc.index === "number" ? tc.index : 0;
-          let buf = toolCalls.get(idx);
-          const incomingName = tc.function?.name ?? null;
-          const incomingId = tc.id ?? null;
-          if (!buf) {
-            buf = { id: incomingId ?? `tc_${idx}`, name: incomingName ?? "", args: "" };
-            toolCalls.set(idx, buf);
-            if (buf.name) {
-              yield { type: "tool_call_start", index: idx, id: buf.id, name: buf.name };
+    if (choice) {
+      const d = choice.delta;
+      if (d) {
+        if (typeof d.reasoning_content === "string" && d.reasoning_content.length) {
+          yield { type: "reasoning", delta: d.reasoning_content };
+        }
+        if (typeof d.content === "string" && d.content.length) {
+          yield { type: "text", delta: d.content };
+        }
+        if (Array.isArray(d.tool_calls)) {
+          for (const tc of d.tool_calls) {
+            const idx = typeof tc.index === "number" ? tc.index : 0;
+            let buf = toolCalls.get(idx);
+            const incomingName = tc.function?.name ?? null;
+            const incomingId = tc.id ?? null;
+            if (!buf) {
+              buf = { id: incomingId ?? `tc_${idx}`, name: incomingName ?? "", args: "" };
+              toolCalls.set(idx, buf);
+              if (buf.name) {
+                yield { type: "tool_call_start", index: idx, id: buf.id, name: buf.name };
+              }
+            } else {
+              if (!buf.name && incomingName) {
+                buf.name = incomingName;
+                yield { type: "tool_call_start", index: idx, id: buf.id, name: buf.name };
+              }
+              if (buf.id.startsWith("tc_") && incomingId) buf.id = incomingId;
             }
-          } else {
-            if (!buf.name && incomingName) {
-              buf.name = incomingName;
-              yield { type: "tool_call_start", index: idx, id: buf.id, name: buf.name };
+            const argDelta = tc.function?.arguments;
+            if (typeof argDelta === "string" && argDelta.length) {
+              buf.args += argDelta;
+              yield { type: "tool_call_args", index: idx, argsDelta: argDelta };
             }
-            if (buf.id.startsWith("tc_") && incomingId) buf.id = incomingId;
-          }
-          const argDelta = tc.function?.arguments;
-          if (typeof argDelta === "string" && argDelta.length) {
-            buf.args += argDelta;
-            yield { type: "tool_call_args", index: idx, argsDelta: argDelta };
           }
         }
       }
+      if (choice.finish_reason) finishReason = choice.finish_reason;
     }
-    if (choice.finish_reason) finishReason = choice.finish_reason;
   }
   for (const [idx, buf] of [...toolCalls.entries()].sort((a, b) => a[0] - b[0])) {
     if (!buf.name) continue;
@@ -965,10 +975,129 @@ function safeJsonParse(text) {
     return null;
   }
 }
-var EXTRACTORS;
+function truncate(str, max) {
+  if (str.length <= max) return str;
+  return str.slice(0, max) + "\u2026";
+}
+async function callLlm(messages, llmOpts, maxTokens = 64) {
+  if (!llmOpts) return "";
+  const events = runKimi({
+    accountId: llmOpts.accountId,
+    apiToken: llmOpts.apiToken,
+    model: llmOpts.model,
+    messages,
+    temperature: 0.1,
+    maxCompletionTokens: maxTokens,
+    gateway: llmOpts.gateway,
+    signal: llmOpts.signal
+  });
+  let text = "";
+  for await (const ev of events) {
+    if (ev.type === "text") text += ev.delta;
+  }
+  return text.trim().replace(/^["']|["']$/g, "").replace(/\s+/g, " ").toLowerCase();
+}
+async function synthesizeEditEvent(file, toolName, toolArgs, assistantMessage, llmOpts) {
+  if (!llmOpts) return null;
+  const oldString = typeof toolArgs.old_string === "string" ? toolArgs.old_string : "";
+  const newString = typeof toolArgs.new_string === "string" ? toolArgs.new_string : "";
+  const fullContent = typeof toolArgs.content === "string" ? toolArgs.content : "";
+  const isWrite = toolName === "write";
+  const before = isWrite ? "(new file)" : truncate(oldString, 600);
+  const after = isWrite ? truncate(fullContent, 600) : truncate(newString, 600);
+  const intent = assistantMessage ? assistantMessage.slice(-300).trim() : "";
+  const changeContext = `File: ${file}
+Tool: ${toolName}
+
+Before:
+${before}
+
+After:
+${after}${intent ? `
+
+Context: ${intent}` : ""}`;
+  const summary1 = await callLlm(
+    [
+      { role: "system", content: EDIT_SYNTHESIS_SYSTEM },
+      { role: "user", content: `${changeContext}
+
+Summary:` }
+    ],
+    llmOpts
+  );
+  if (summary1.length >= 10 && summary1.length <= 200) {
+    const verdict = await callLlm(
+      [
+        { role: "system", content: VERIFIER_SYSTEM },
+        {
+          role: "user",
+          content: `${changeContext}
+
+Proposed summary: "${summary1}"
+
+Is this accurate?`
+        }
+      ],
+      llmOpts,
+      8
+    );
+    if (verdict.startsWith("yes")) return summary1;
+  }
+  const retrySystem = `${EDIT_SYNTHESIS_SYSTEM}
+
+CRITICAL: The previous summary was rejected. Be specific. Include concrete details from the After section.`;
+  const summary2 = await callLlm(
+    [
+      { role: "system", content: retrySystem },
+      { role: "user", content: `${changeContext}
+
+Summary:` }
+    ],
+    llmOpts
+  );
+  if (summary2.length >= 10 && summary2.length <= 200) {
+    const verdict2 = await callLlm(
+      [
+        { role: "system", content: VERIFIER_SYSTEM },
+        {
+          role: "user",
+          content: `${changeContext}
+
+Proposed summary: "${summary2}"
+
+Is this accurate?`
+        }
+      ],
+      llmOpts,
+      8
+    );
+    if (verdict2.startsWith("yes")) return summary2;
+  }
+  return null;
+}
+var EDIT_SYNTHESIS_SYSTEM, VERIFIER_SYSTEM, EXTRACTORS;
 var init_extractors = __esm({
   "src/memory/extractors.ts"() {
     "use strict";
+    init_client();
+    EDIT_SYNTHESIS_SYSTEM = `You summarize a SINGLE code edit. Write ONE concise sentence (max 20 words) describing exactly what changed.
+
+Rules:
+- Use ONLY the Before/After diff below.
+- For new files: describe the file's content or purpose. Never say just "added a new file".
+- For edits: describe the specific code change.
+
+Examples:
+- Created test-memory.md containing the text "Memory test".
+- Fixed race condition in loop.ts by adding AbortSignal guard.
+- Added vitest dependency and removed jest from package.json.
+
+Respond with only the summary sentence. No quotes, no preamble.`;
+    VERIFIER_SYSTEM = `You verify whether a summary accurately describes a code edit.
+Answer exactly "yes" if the summary correctly captures what changed in the file.
+Answer exactly "no" if the summary is vague, wrong, or misses the actual change.
+
+Respond with only "yes" or "no".`;
     EXTRACTORS = [
       {
         id: "package_json",
@@ -1022,9 +1151,27 @@ var init_extractors = __esm({
       {
         id: "edit_event",
         match: (tool, file) => (tool === "edit" || tool === "write") && !!file,
-        extract: (_content, file) => {
+        extract: async (_content, file, ctx) => {
           if (!file) return null;
           const safeKey = file.replace(/[^a-zA-Z0-9]/g, "_");
+          if (ctx?.llmOpts && (ctx.toolArgs || ctx.assistantMessage)) {
+            const summary = await synthesizeEditEvent(
+              file,
+              ctx.toolArgs?._toolName || "edit",
+              ctx.toolArgs || {},
+              ctx.assistantMessage,
+              ctx.llmOpts
+            );
+            if (summary) {
+              return {
+                content: summary,
+                category: "event",
+                importance: 3,
+                topicKey: `event_edit_${safeKey}`,
+                relatedFiles: [file]
+              };
+            }
+          }
           return {
             content: `File modified: ${file}.`,
             category: "event",
@@ -1715,25 +1862,44 @@ ${sandboxResult.output}` : sandboxResult.output;
         opts2.callbacks.onToolResult?.(result);
         if (opts2.memoryManager) {
           let filePath;
+          let toolArgs = {};
           try {
-            const args = JSON.parse(tc.function.arguments || "{}");
-            filePath = args.path;
+            toolArgs = JSON.parse(tc.function.arguments || "{}");
+            filePath = toolArgs.path;
           } catch {
           }
+          const lastAssistant = [...opts2.messages].reverse().find(
+            (m) => m.role === "assistant" && m.tool_calls && m.tool_calls.length > 0
+          );
+          const assistantMessage = lastAssistant?.content ?? "";
+          const llmOpts = opts2.memoryManager.getExtractionLlmOpts();
           for (const extractor of EXTRACTORS) {
             if (extractor.match(tc.function.name, filePath)) {
-              const memory = extractor.extract(result.content, filePath);
-              if (memory) {
-                void opts2.memoryManager.remember(
-                  memory.content,
-                  memory.category,
-                  memory.importance,
-                  opts2.cwd,
-                  opts2.sessionId ?? "unknown",
-                  opts2.signal
-                ).catch(() => {
-                });
-              }
+              void (async () => {
+                try {
+                  const memory = await extractor.extract(result.content, filePath, {
+                    toolArgs: { ...toolArgs, _toolName: tc.function.name },
+                    assistantMessage: typeof assistantMessage === "string" ? assistantMessage : "",
+                    llmOpts: {
+                      ...llmOpts,
+                      signal: opts2.signal
+                    }
+                  });
+                  if (memory) {
+                    await opts2.memoryManager.remember(
+                      memory.content,
+                      memory.category,
+                      memory.importance,
+                      opts2.cwd,
+                      opts2.sessionId ?? "unknown",
+                      opts2.signal,
+                      void 0,
+                      memory.topicKey
+                    );
+                  }
+                } catch {
+                }
+              })();
             }
           }
         }
@@ -9392,6 +9558,18 @@ Return a JSON array of strings. Example:
           gateway: this.opts.gateway
         };
       }
+      get extractionLlmOpts() {
+        return {
+          accountId: this.opts.accountId,
+          apiToken: this.opts.apiToken,
+          model: this.opts.extractionModel ?? "@cf/meta/llama-3.2-3b-instruct",
+          gateway: this.opts.gateway
+        };
+      }
+      /** Expose extraction LLM opts so the agent loop can pass them to extractors. */
+      getExtractionLlmOpts() {
+        return this.extractionLlmOpts;
+      }
       shouldRedact() {
         return this.opts.redactSecrets !== false;
       }
@@ -9399,7 +9577,7 @@ Return a JSON array of strings. Example:
        * Store a memory with verification, topic-key normalization, hypothetical queries,
        * secret redaction, and supersession.
        */
-      async remember(content, category, importance, repoPath, sessionId, signal, agentRole) {
+      async remember(content, category, importance, repoPath, sessionId, signal, agentRole, topicKey) {
         if (!this.db) throw new Error("Memory DB not open");
         let safeContent = this.shouldRedact() ? redactSecrets(content) : content;
         if (!safeContent.trim()) {
@@ -9412,10 +9590,10 @@ Return a JSON array of strings. Example:
         if (verified.corrected_content) {
           safeContent = verified.corrected_content;
         }
-        const topicKey = this.normalizeTopicKey(safeContent, repoPath);
+        const resolvedTopicKey = topicKey?.trim() || this.normalizeTopicKey(safeContent, repoPath);
         const supersededIds = [];
-        if (topicKey) {
-          const existing = findMemoriesByTopicKey(this.db, repoPath, topicKey);
+        if (resolvedTopicKey) {
+          const existing = findMemoriesByTopicKey(this.db, repoPath, resolvedTopicKey);
           for (const old of existing) {
             supersedeMemory(this.db, old.id, "pending");
             supersededIds.push(old.id);
@@ -9436,7 +9614,7 @@ Return a JSON array of strings. Example:
           sourceSessionId: sessionId,
           repoPath,
           importance: Math.max(1, Math.min(5, importance)),
-          topicKey: topicKey ?? void 0,
+          topicKey: resolvedTopicKey ?? void 0,
           agentRole
         };
         const memory = insertMemory(this.db, input, embeddings[0]);
@@ -12115,6 +12293,7 @@ function App({
         apiToken: cfg.apiToken,
         model: cfg.model,
         plumbingModel: cfg.plumbingModel,
+        extractionModel: cfg.memoryExtractionModel,
         embeddingModel: cfg.memoryEmbeddingModel,
         gateway: gatewayFromConfig(cfg),
         maxAgeDays: cfg.memoryMaxAgeDays ?? RETENTION.memoryMaxAgeDays,
@@ -12807,8 +12986,6 @@ function App({
         sessionId: ensureSessionId(),
         memoryManager: memoryManagerRef.current,
         codeMode: effectiveCodeMode,
-        maxInputTokens: effectiveCodeMode ? 2e5 : void 0,
-        continueOnLimit: effectiveCodeMode ? true : void 0,
         onIterationEnd,
         onFileChange: (path, content) => {
           if (content) {
@@ -14008,8 +14185,6 @@ ${lines.join("\n")}` }]);
           memoryManager: memoryManagerRef.current,
           keepLastImageTurns: cfg.imageHistoryTurns ?? 2,
           codeMode: effectiveCodeMode,
-          maxInputTokens: effectiveCodeMode ? 2e5 : void 0,
-          continueOnLimit: effectiveCodeMode ? true : void 0,
           onIterationEnd,
           intentClassification: classification,
           onFileChange: (path, content2) => {