kimiflare 0.13.0 → 0.13.1

package/dist/index.js

@@ -230,57 +230,87 @@ async function* parseStream(body, signal) {
   const toolCalls = /* @__PURE__ */ new Map();
   let lastUsage = null;
   let finishReason = null;
-  for await (const dataStr of readSSE(body, signal)) {
-    if (dataStr === "[DONE]") break;
-    let chunk = null;
-    try {
-      chunk = JSON.parse(dataStr);
-    } catch {
-      continue;
-    }
-    if (!chunk) continue;
-    if (chunk.usage) {
-      lastUsage = chunk.usage;
-      yield { type: "usage", usage: chunk.usage };
-    }
-    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 };
+  let timeoutId = null;
+  let timedOut = false;
+  const resetTimeout = () => {
+    if (timeoutId) clearTimeout(timeoutId);
+    timeoutId = setTimeout(() => {
+      timedOut = true;
+    }, STREAM_TIMEOUT_MS);
+  };
+  try {
+    for await (const dataStr of readSSE(body, signal)) {
+      if (timedOut) {
+        throw new KimiApiError(
+          `kimiflare: stream timed out (no data for ${STREAM_TIMEOUT_MS / 1e3}s)`,
+          void 0,
+          void 0
+        );
       }
-      if (typeof d.content === "string" && d.content.length) {
-        yield { type: "text", delta: d.content };
+      resetTimeout();
+      if (dataStr === "[DONE]") break;
+      let chunk = null;
+      try {
+        chunk = JSON.parse(dataStr);
+      } catch {
+        continue;
       }
-      if (Array.isArray(d.tool_calls)) {
-        for (const tc of d.tool_calls) {
-          const idx = typeof tc.index === "number" ? tc.index : 0;
-          let buf = toolCalls.get(idx);
-          const incomingName = tc.function?.name ?? null;
-          const incomingId = tc.id ?? null;
-          if (!buf) {
-            buf = { id: incomingId ?? `tc_${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 (!chunk) continue;
+      if (chunk.usage) {
+        lastUsage = chunk.usage;
+        yield { type: "usage", usage: chunk.usage };
+      }
+      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 };
+              }
+            } 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;
+      yield {
+        type: "tool_call_complete",
+        index: idx,
+        id: buf.id,
+        name: buf.name,
+        arguments: buf.args
+      };
+    }
+  } finally {
+    if (timeoutId) clearTimeout(timeoutId);
   }
   for (const [idx, buf] of [...toolCalls.entries()].sort((a, b) => a[0] - b[0])) {
     if (!buf.name) continue;
@@ -354,7 +384,7 @@ function sleep(ms, signal) {
     signal?.addEventListener("abort", onAbort, { once: true });
   });
 }
-var RETRYABLE_CODES, MAX_ATTEMPTS;
+var RETRYABLE_CODES, MAX_ATTEMPTS, STREAM_TIMEOUT_MS;
 var init_client = __esm({
   "src/agent/client.ts"() {
     "use strict";
@@ -363,6 +393,7 @@ var init_client = __esm({
     init_messages();
     RETRYABLE_CODES = /* @__PURE__ */ new Set([3040]);
     MAX_ATTEMPTS = 5;
+    STREAM_TIMEOUT_MS = 6e4;
   }
 });
 
@@ -3575,6 +3606,60 @@ function findImagePaths(text) {
   }
   return [...new Set(paths)];
 }
+function stripImagesFromHistory(messages) {
+  for (const m of messages) {
+    if (!Array.isArray(m.content)) continue;
+    let changed = false;
+    const next = [];
+    for (const part of m.content) {
+      if (part.type === "image_url") {
+        changed = true;
+        next.push({ type: "text", text: "[image]" });
+      } else {
+        next.push(part);
+      }
+    }
+    if (changed) {
+      m.content = next;
+    }
+  }
+}
+function truncateOldToolResults(messages, keepRecent) {
+  const toolIndices = [];
+  for (let i = 0; i < messages.length; i++) {
+    if (messages[i].role === "tool") toolIndices.push(i);
+  }
+  const cutoff = toolIndices.length - keepRecent;
+  for (let i = 0; i < cutoff; i++) {
+    const idx = toolIndices[i];
+    const m = messages[idx];
+    const text = typeof m.content === "string" ? m.content : "";
+    if (text.length > 500) {
+      m.content = text.slice(0, 500) + "\n\u2026 [truncated]";
+    }
+  }
+}
+async function maybeAutoCompact(messages, cfg, onInfo) {
+  const heapUsed = process.memoryUsage().heapUsed;
+  if (heapUsed < HEAP_AUTO_COMPACT_THRESHOLD || messages.length <= MSG_AUTO_COMPACT_THRESHOLD) {
+    return false;
+  }
+  onInfo("auto-compacting to reduce memory usage");
+  const result = await compactMessages({
+    accountId: cfg.accountId,
+    apiToken: cfg.apiToken,
+    model: cfg.model,
+    messages,
+    keepLastTurns: 2
+  });
+  if (result.replacedCount > 0) {
+    messages.length = 0;
+    messages.push(...result.newMessages);
+    onInfo(`compacted ${result.replacedCount} messages into a summary`);
+    return true;
+  }
+  return false;
+}
 function App({ initialCfg, initialUpdateResult }) {
   const { exit } = useApp();
   const [cfg, setCfg] = useState6(initialCfg);
@@ -4002,6 +4087,13 @@ function App({ initialCfg, initialUpdateResult }) {
           })
         }
       });
+      stripImagesFromHistory(messagesRef.current);
+      truncateOldToolResults(messagesRef.current, 8);
+      await maybeAutoCompact(
+        messagesRef.current,
+        cfg,
+        (text) => setEvents((es) => [...es, { kind: "info", key: mkKey(), text }])
+      );
       if (existsSync(join5(cwd, "KIMI.md"))) {
         messagesRef.current[0] = {
           role: "system",
@@ -4456,7 +4548,14 @@ use: /thinking low | medium | high`
             })
           }
         });
-        await saveSessionSafe();
+        stripImagesFromHistory(messagesRef.current);
+        truncateOldToolResults(messagesRef.current, 8);
+        await maybeAutoCompact(
+          messagesRef.current,
+          cfg,
+          (text2) => setEvents((es) => [...es, { kind: "info", key: mkKey(), text: text2 }])
+        );
+        void saveSessionSafe();
       } catch (e) {
         if (e.name === "AbortError") {
           setEvents((es) => [...es, { kind: "info", key: mkKey(), text: "(interrupted)" }]);
@@ -4652,7 +4751,7 @@ async function renderApp(cfg, updateResult) {
   const instance = render(/* @__PURE__ */ jsx13(App, { initialCfg: cfg, initialUpdateResult: updateResult }));
   await instance.waitUntilExit();
 }
-var CONTEXT_LIMIT, AUTO_COMPACT_SUGGEST_PCT, MAX_EVENTS, nextAssistantId, nextKey, mkKey, MAX_IMAGES_PER_MESSAGE, EFFORT_DESCRIPTIONS;
+var CONTEXT_LIMIT, AUTO_COMPACT_SUGGEST_PCT, MAX_EVENTS, nextAssistantId, nextKey, mkKey, MAX_IMAGES_PER_MESSAGE, EFFORT_DESCRIPTIONS, HEAP_AUTO_COMPACT_THRESHOLD, MSG_AUTO_COMPACT_THRESHOLD;
 var init_app = __esm({
   "src/app.tsx"() {
     "use strict";
@@ -4689,6 +4788,8 @@ var init_app = __esm({
       medium: "medium \u2014 balanced (default). Solid quality on most edits, fast on trivial prompts.",
       high: "high \u2014 deepest reasoning; slowest. Best for complex debugging, architecture, multi-file refactors."
     };
+    HEAP_AUTO_COMPACT_THRESHOLD = 2e9;
+    MSG_AUTO_COMPACT_THRESHOLD = 50;
   }
 });