kimiflare 0.88.1 → 0.88.3

package/dist/index.js

@@ -3418,15 +3418,87 @@ async function fetchWithRetry(url, init, retries = 3) {
   }
   throw lastError ?? new Error("embeddings request failed after retries");
 }
+function parseOpenAiEmbeddingResponse(json2) {
+  if (!json2 || typeof json2 !== "object") {
+    throw new Error("embeddings response was not an object");
+  }
+  const data = json2.data;
+  if (!Array.isArray(data)) {
+    throw new Error("embeddings response contained no data array");
+  }
+  const indexed = [];
+  for (const item of data) {
+    if (!item || typeof item !== "object") continue;
+    const embedding = item.embedding;
+    const idx = item.index;
+    if (!Array.isArray(embedding)) continue;
+    indexed.push({
+      index: typeof idx === "number" ? idx : indexed.length,
+      vector: new Float32Array(embedding)
+    });
+  }
+  if (indexed.length === 0) {
+    throw new Error("embeddings response contained no vectors");
+  }
+  indexed.sort((a, b) => a.index - b.index);
+  return indexed.map((item) => {
+    if (item.vector.length === 0) {
+      throw new Error("embeddings response contained empty vector");
+    }
+    return item.vector;
+  });
+}
+function parseWorkersAiEmbeddingResponse(json2) {
+  let vectors = [];
+  if (json2 && typeof json2 === "object") {
+    const result = json2.result;
+    if (result && typeof result === "object") {
+      const data = result.data;
+      if (Array.isArray(data)) {
+        if (Array.isArray(data[0])) {
+          vectors = data;
+        } else {
+          const shape = result.shape;
+          if (shape && shape.length === 2) {
+            const dim = shape[1];
+            const flat = data;
+            vectors = [];
+            for (let i = 0; i < flat.length; i += dim) {
+              vectors.push(flat.slice(i, i + dim));
+            }
+          }
+        }
+      }
+    }
+  }
+  if (vectors.length === 0) {
+    throw new Error("embeddings response contained no vectors");
+  }
+  return vectors.map((vec) => {
+    const arr = new Float32Array(vec);
+    if (arr.length === 0) {
+      throw new Error("embeddings response contained empty vector");
+    }
+    return arr;
+  });
+}
 async function fetchEmbeddings(opts2) {
   const model = opts2.model ?? DEFAULT_MODEL2;
+  const texts = opts2.texts.map(truncateForEmbedding);
+  if (texts.length === 0) {
+    return [];
+  }
   let url;
   const headers = {
     "Content-Type": "application/json",
     "User-Agent": getUserAgent()
   };
+  let body;
+  let parseResponse;
   if (opts2.gateway) {
-    url = `https://gateway.ai.cloudflare.com/v1/${opts2.accountId}/${opts2.gateway.id}/workers-ai/${model}`;
+    url = `https://gateway.ai.cloudflare.com/v1/${encodeURIComponent(
+      opts2.accountId
+    )}/${encodeURIComponent(opts2.gateway.id)}/compat/embeddings`;
     headers.Authorization = `Bearer ${opts2.apiToken}`;
     const merged = {
       ...opts2.gateway.metadata ?? {},
@@ -3440,48 +3512,22 @@ async function fetchEmbeddings(opts2) {
     if (opts2.gateway.skipCache !== void 0) {
       headers["cf-aig-skip-cache"] = String(opts2.gateway.skipCache);
     }
+    body = JSON.stringify({
+      model: `workers-ai/${model}`,
+      input: texts
+    });
+    parseResponse = parseOpenAiEmbeddingResponse;
   } else {
-    url = `https://api.cloudflare.com/client/v4/accounts/${opts2.accountId}/ai/run/${model}`;
+    url = `https://api.cloudflare.com/client/v4/accounts/${encodeURIComponent(
+      opts2.accountId
+    )}/ai/run/${encodeURIComponent(model)}`;
     headers.Authorization = `Bearer ${opts2.apiToken}`;
+    body = JSON.stringify({ text: texts });
+    parseResponse = parseWorkersAiEmbeddingResponse;
   }
-  const results = [];
-  for (const text of opts2.texts) {
-    const truncated = truncateForEmbedding(text);
-    const body = JSON.stringify({ text: [truncated] });
-    const res = await fetchWithRetry(url, { method: "POST", headers, body });
-    const json2 = await res.json();
-    let vectors = [];
-    if (json2 && typeof json2 === "object") {
-      const result = json2.result;
-      if (result && typeof result === "object") {
-        const data = result.data;
-        if (Array.isArray(data)) {
-          if (Array.isArray(data[0])) {
-            vectors = data;
-          } else {
-            const shape = result.shape;
-            if (shape && shape.length === 2) {
-              const dim = shape[1];
-              const flat = data;
-              vectors = [];
-              for (let i = 0; i < flat.length; i += dim) {
-                vectors.push(flat.slice(i, i + dim));
-              }
-            }
-          }
-        }
-      }
-    }
-    if (vectors.length === 0) {
-      throw new Error("embeddings response contained no vectors");
-    }
-    const vec = new Float32Array(vectors[0]);
-    if (vec.length === 0) {
-      throw new Error("embeddings response contained empty vector");
-    }
-    results.push(vec);
-  }
-  return results;
+  const res = await fetchWithRetry(url, { method: "POST", headers, body });
+  const json2 = await res.json();
+  return parseResponse(json2);
 }
 function cosineSimilarity(a, b) {
   if (a.length !== b.length) {
@@ -11191,6 +11237,50 @@ Return a JSON array of strings. Example:
         }
         return { id: memory.id, superseded: supersededIds.length > 0 ? supersededIds : void 0 };
       }
+      /**
+       * Store a plan directly under a deterministic topic key.
+       * Skips embedding, verification, and hypothetical queries so it is fast and
+       * deterministic. Supersedes any previous plan stored under the same key.
+       */
+      async rememberPlan(plan, repoPath, sessionId, topicKey = "current_dev_plan") {
+        if (!this.db) throw new Error("Memory DB not open");
+        const safeContent = this.shouldRedact() ? redactSecrets(plan) : plan;
+        if (!safeContent.trim()) {
+          throw new Error("Plan content is empty after redaction");
+        }
+        const normalizedKey = topicKey.trim();
+        if (!normalizedKey) {
+          throw new Error("Plan topic key cannot be empty");
+        }
+        const supersededIds = [];
+        const existing = findMemoriesByTopicKey(this.db, repoPath, normalizedKey);
+        for (const old of existing) {
+          supersedeMemory(this.db, old.id, "pending");
+          supersededIds.push(old.id);
+        }
+        const zeroEmbedding = new Float32Array(DEFAULT_EMBEDDING_DIM);
+        const memory = insertMemory(this.db, {
+          content: safeContent,
+          category: "task",
+          sourceSessionId: sessionId,
+          repoPath,
+          importance: 4,
+          topicKey: normalizedKey
+        }, zeroEmbedding);
+        for (const oldId of supersededIds) {
+          supersedeMemory(this.db, oldId, memory.id);
+        }
+        return { id: memory.id, superseded: supersededIds.length > 0 ? supersededIds : void 0 };
+      }
+      /**
+       * Recall the latest memory for an exact topic key.
+       * Does not use embeddings; returns null if no matching memory exists.
+       */
+      getByTopicKey(repoPath, topicKey) {
+        if (!this.db) return null;
+        const rows = findMemoriesByTopicKey(this.db, repoPath, topicKey);
+        return rows[0] ?? null;
+      }
       /**
        * Count high-signal memories created since the given timestamp.
        * Used for KIMI.md drift detection (Trigger A: session-start check).
@@ -16777,6 +16867,32 @@ var init_distill = __esm({
   }
 });
 
+// src/agent/plan-resolver.ts
+function resolvePlanForFresh(opts2) {
+  const { mode, messages, sessionPlan, memoryManager, memoryEnabled, repoPath } = opts2;
+  if (mode !== "plan") {
+    return null;
+  }
+  if (sessionPlan) {
+    return sessionPlan;
+  }
+  if (memoryEnabled && memoryManager) {
+    const stored = memoryManager.getByTopicKey(repoPath, PLAN_MEMORY_TOPIC_KEY);
+    if (stored?.content) {
+      return stored.content;
+    }
+  }
+  return distillSessionPlan(messages);
+}
+var PLAN_MEMORY_TOPIC_KEY;
+var init_plan_resolver = __esm({
+  "src/agent/plan-resolver.ts"() {
+    "use strict";
+    init_distill();
+    PLAN_MEMORY_TOPIC_KEY = "current_dev_plan";
+  }
+});
+
 // src/agent/continuation-summary.ts
 import { execSync as execSync4 } from "child_process";
 function extractFirstUserGoal(messages) {
@@ -19863,6 +19979,10 @@ Executor opened PR: ${prUrl}` : plan });
         const plan = distillSessionPlan(messages);
         if (plan) {
           sessionPlan = plan;
+          if (startupCfg?.memoryEnabled && memoryManager) {
+            void memoryManager.rememberPlan(plan, process.cwd(), randomUUID2()).catch(() => {
+            });
+          }
         }
       }
       if (planOptionsRef.current && !currentController?.signal.aborted) {
@@ -21609,7 +21729,14 @@ changelog-image failed: ${err instanceof Error ? err.message : String(err)}
           return true;
         }
         void (async () => {
-          const summary = await generateContinuationSummary({
+          const summary = currentMode === "plan" ? resolvePlanForFresh({
+            mode: currentMode,
+            messages,
+            sessionPlan,
+            memoryManager,
+            memoryEnabled: startupCfg?.memoryEnabled,
+            repoPath: process.cwd()
+          }) : await generateContinuationSummary({
             messages,
             mode: currentMode,
             accountId: opts2.accountId,
@@ -21908,6 +22035,7 @@ var init_ui_mode = __esm({
     init_sessions();
     init_llm_summarize();
     init_distill();
+    init_plan_resolver();
     init_continuation_summary();
     init_greetings();
     init_theme();
@@ -31461,6 +31589,7 @@ var init_slash_commands = __esm({
     init_session_store();
     init_deploy();
     init_tui_auth();
+    init_plan_resolver();
     init_continuation_summary();
     init_clipboard();
     handleExit = (ctx) => {
@@ -31513,7 +31642,14 @@ var init_slash_commands = __esm({
         ]);
         return true;
       }
-      const summary = await generateContinuationSummary({
+      const summary = ctx.mode === "plan" ? resolvePlanForFresh({
+        mode: ctx.mode,
+        messages: ctx.messagesRef.current,
+        sessionPlan: ctx.sessionPlanRef.current,
+        memoryManager: ctx.memoryManagerRef.current,
+        memoryEnabled: cfg?.memoryEnabled,
+        repoPath: process.cwd()
+      }) : await generateContinuationSummary({
         messages: ctx.messagesRef.current,
         mode: ctx.mode,
         accountId: cfg?.accountId ?? "",
@@ -34953,7 +35089,14 @@ ${wcagWarnings.join("\n")}` }
     (picked) => {
       setShowPlanCompletePicker(false);
       if (!picked || picked === "continue") return;
-      const plan = sessionPlanRef.current ?? distillSessionPlan(messagesRef.current);
+      const plan = resolvePlanForFresh({
+        mode: "plan",
+        messages: messagesRef.current,
+        sessionPlan: sessionPlanRef.current,
+        memoryManager: memoryManagerRef.current,
+        memoryEnabled: cfg?.memoryEnabled,
+        repoPath: process.cwd()
+      });
       if (!plan) {
         setEvents((e) => [
           ...e,
@@ -35657,6 +35800,10 @@ ${conflicts.join("\n")}` }
               const plan = distillSessionPlan(messagesRef.current);
               if (plan) {
                 sessionPlanRef.current = plan;
+                if (cfg?.memoryEnabled && memoryManagerRef.current && sessionIdRef.current) {
+                  void memoryManagerRef.current.rememberPlan(plan, process.cwd(), sessionIdRef.current).catch(() => {
+                  });
+                }
                 setShowPlanCompletePicker(true);
               }
             }
@@ -36245,6 +36392,7 @@ var init_app = __esm({
     init_manager_init();
     init_run_compact();
     init_distill();
+    init_plan_resolver();
     init_command_handlers();
     init_app_helpers();
   }