pi-freeflow 1.44.1 → 1.45.0

package/src/daemon.ts

@@ -10,6 +10,7 @@
  */
 
 import path from "node:path";
+import fs from "node:fs";
 import { fileURLToPath } from "node:url";
 import {
  DAEMON_GC_MS,
@@ -34,12 +35,26 @@ import {
 import { getActiveRequests, startClineKeepAlive, startProxy } from "./proxy.ts";
 import { loadRelayState, setActiveRelayState } from "./relay-state.ts";
 
+/** True when this module was run as the entry script (not imported by tests). */
+function safeCanonicalPath(p: string): string {
+ try {
+  return fs.realpathSync.native(path.resolve(p));
+ } catch {
+  return path.resolve(p);
+ }
+}
+
 /** True when this module was run as the entry script (not imported by tests). */
 export function isDaemonMain(): boolean {
  try {
   const entry = process.argv[1];
   if (!entry) return false;
-  return path.resolve(entry) === fileURLToPath(import.meta.url);
+  const canonicalEntry = safeCanonicalPath(entry);
+  const canonicalSelf = safeCanonicalPath(fileURLToPath(import.meta.url));
+  if (process.platform === "win32") {
+   return canonicalEntry.toLowerCase() === canonicalSelf.toLowerCase();
+  }
+  return canonicalEntry === canonicalSelf;
  } catch {
   return false;
  }

package/src/deploy.ts

@@ -80,6 +80,7 @@ const isPrivateHostname = function(h) {
     return false
   }
   if (host.includes(":")) {
+    if (host.startsWith("64:ff9b:") || host.startsWith("2002:")) return true
     if (host.startsWith("fc") || host.startsWith("fd")) return true
     if (/^fe[89ab]/.test(host)) return true
     return false
@@ -121,7 +122,7 @@ async function relayHandler(req) {
   DENY_HEADERS.forEach((h) => headers.delete(h));
   headers.delete("x-relay-auth");
   try {
-    const init = { method: req.method, headers };
+    const init = { method: req.method, headers, redirect: "manual" };
     if (req.method !== "GET" && req.method !== "HEAD") { init.body = req.body; init.duplex = "half"; }
     const response = await fetch(resolved.url, init);
     return new Response(response.body, { status: response.status, headers: response.headers });
@@ -242,6 +243,7 @@ export async function deployVercelRelay(
 
  const depJson = (await dep.json()) as { id?: string; uid?: string; projectId?: string };
  const depId = depJson.id || depJson.uid;
+ if (!depId) throw new Error("Vercel did not return a deployment id");
  const projectId = depJson.projectId || name;
 
  // 2. Make the deployment public (disable SSO protection if enabled on team)

package/src/fingerprint-tools.ts

@@ -41,7 +41,7 @@ export function clampedEffortFor(model: string, effort: string): string | null {
  // Alias of a supported level: rewrite, because upstream knows the level, not
  // the alias.
  if (isUltra && key in map && map[key] !== null && map[key] !== undefined) return map[key];
- // Effort is supported (non-null) → no clamp
+ // Effort is supported (non-null) → no clamp needed
  if (key in map && map[key] !== null && map[key] !== undefined) return null;
  // Effort key is unknown and not "ultra" → let upstream decide
  if (!(key in map) && !isUltra) return null;

package/src/index.ts

@@ -16,7 +16,7 @@ import {
  refreshCatalog,
  setAliveCatalog,
 } from "./catalog.ts";
-import { ensureDaemon as ensureClientDaemon, ensureProxyReady, getClientPort, hasFallbackServer, stopHeartbeat } from "./client.ts";
+import { ensureDaemon as ensureClientDaemon, ensureProxyReady, getClientPort, stopHeartbeat } from "./client.ts";
 import { createCommandSpec, stopLogsFollow, updateStatusBar } from "./commands.ts";
 import { HOST, ONBOARDED_FLAG_FILE, PORT } from "./config.ts";
 import { logInfo, logWarn } from "./logger.ts";
@@ -179,6 +179,9 @@ export function buildProviderConfig(
          supportsDeveloperRole: false,
          supportsReasoningEffort: true,
         }),
+     ...(m.thinkingLevelMap !== undefined
+      ? { reasoningEffortMap: Object.fromEntries(Object.entries(m.thinkingLevelMap).filter(([, v]) => typeof v === "string")) }
+      : {}),
      ...(m.requiresReasoningContentOnAssistantMessages !== undefined
       ? { requiresReasoningContentOnAssistantMessages: m.requiresReasoningContentOnAssistantMessages }
       : {}),
@@ -371,15 +374,16 @@ export default async function(pi: ExtensionAPI): Promise<void> {
    ctx.ui?.setStatus?.("freeflow", undefined);
   }
  });
- pi.on?.("session_shutdown", () => {
+ pi.on?.("session_shutdown", async () => {
   stopLogsFollow();
-  if (hasFallbackServer()) {
-   stopHeartbeat();
-  }
+  await stopHeartbeat();
+ });
+ process.once("beforeExit", () => {
+  void stopHeartbeat();
  });
  process.once("exit", () => {
   try {
-   stopHeartbeat();
+   void stopHeartbeat();
   } catch { }
  });
 }

package/src/lane-translation.ts

@@ -49,10 +49,20 @@ export function responsesInputToChatMessages(input: unknown): Record<string, unk
   if (typeof item !== "object" || item === null || Array.isArray(item)) continue;
   const rec = item as Record<string, unknown>;
   if (rec.type === "function_call") {
-   const args = typeof rec.arguments === "string" ? rec.arguments : JSON.stringify(rec.arguments ?? {});
+   const args = typeof rec.arguments === "string" ? rec.arguments : (rec.arguments !== undefined && rec.arguments !== null ? JSON.stringify(rec.arguments) : "{}");
    const name = typeof rec.name === "string" ? rec.name : "tool";
    const id = typeof rec.call_id === "string" ? rec.call_id : typeof rec.id === "string" ? rec.id : "";
-   messages.push({ role: "assistant", content: null, tool_calls: [{ id, type: "function", function: { name, arguments: args } }] });
+   const call = { id, type: "function", function: { name, arguments: args } };
+   const lastMsg = messages[messages.length - 1];
+   if (lastMsg && lastMsg.role === "assistant") {
+    if (Array.isArray(lastMsg.tool_calls)) {
+     (lastMsg.tool_calls as unknown[]).push(call);
+    } else {
+     lastMsg.tool_calls = [call];
+    }
+   } else {
+    messages.push({ role: "assistant", content: null, tool_calls: [call] });
+   }
    continue;
   }
   if (rec.type === "function_call_output") {
@@ -98,6 +108,15 @@ export function chatBodyFromResponsesBody(body: Record<string, unknown>): Record
  for (const key of ["temperature", "top_p", "max_tokens", "max_completion_tokens", "max_output_tokens", "stop", "presence_penalty", "frequency_penalty", "seed", "user", "parallel_tool_calls", "response_format"] as const) {
   if (body[key] !== undefined) out[key] = body[key];
  }
+ if (body.max_output_tokens !== undefined && out.max_tokens === undefined) {
+  out.max_tokens = body.max_output_tokens;
+ }
+ if (out.tool_choice !== undefined && typeof out.tool_choice === "object" && out.tool_choice !== null && !Array.isArray(out.tool_choice)) {
+  const tc = out.tool_choice as Record<string, unknown>;
+  if (tc.type === "function" && typeof tc.name === "string" && tc.function === undefined) {
+   out.tool_choice = { type: "function", function: { name: tc.name } };
+  }
+ }
  return out;
 }
 
@@ -141,7 +160,7 @@ export function chatResponsesJsonFromChatCompletion(chat: Record<string, unknown
     if (typeof tc !== "object" || tc === null || Array.isArray(tc)) continue;
     const rec = tc as Record<string, unknown>;
     const fn = typeof rec.function === "object" && rec.function !== null && !Array.isArray(rec.function) ? (rec.function as Record<string, unknown>) : {};
-    const args = typeof fn.arguments === "string" ? fn.arguments : JSON.stringify(fn.arguments ?? {});
+    const args = typeof fn.arguments === "string" ? fn.arguments : (fn.arguments !== undefined && fn.arguments !== null ? JSON.stringify(fn.arguments) : "{}");
     const callId = typeof rec.id === "string" && rec.id.length > 0 ? rec.id : `call_cline_${output.length}`;
     output.push({ type: "function_call", id: callId, call_id: callId, name: typeof fn.name === "string" ? fn.name : "tool", arguments: args });
    }

package/src/models.ts

@@ -1,9 +1,9 @@
 /**
  * Static model definitions and upstream routing catalogs for pi-freeflow
  *
- * Defines the 32 verified free models (live-verified 2026-10-26):
- * - 8 OpenCode Zen models (all Chat Completions)
- * - 15 KiloCode Keyless Gateway models (OpenRouter format)
+ * Defines the 30 verified free models (live-verified 2026-11-09):
+ * - 11 OpenCode Zen models (all Chat Completions / Responses)
+ * - 10 KiloCode Keyless Gateway models (OpenRouter format)
  * - 1 Cline direct-only model (per-user pool; z-ai/glm-5.3-flash)
  * - 8 OVHcloud AI Endpoints models (keyless, direct-only)
  *
@@ -273,6 +273,64 @@ export const OPENCODE_MODELS: ModelDef[] = [
    max: "max",
   },
  },
+ {
+  // Added 2026-11-03: Stealth reasoning model on OpenCode Zen (/zen/v1/models: crouton-alpha-free).
+  // 1,048,576 context, 270,000 max output. Modalities: text and image.
+  id: "crouton-alpha-free",
+  name: "Crouton Alpha [OpenCode]",
+  reasoning: true,
+  contextWindow: 1_048_576,
+  maxTokens: 270_000,
+  input: ["text", "image"],
+  thinkingLevelMap: {
+   off: null,
+   minimal: "minimal",
+   low: "low",
+   medium: "medium",
+   high: "high",
+   xhigh: "xhigh",
+   max: "max",
+  },
+ },
+ {
+  // Added 2026-11-06: Stealth model on OpenCode Zen (/zen/v1/models: iceman-free).
+  // 262,144 context, 131,072 max output. Modalities: text.
+  id: "iceman-free",
+  name: "Iceman Free [OpenCode]",
+  reasoning: true,
+  contextWindow: 262_144,
+  maxTokens: 131_072,
+  input: ["text"],
+  thinkingLevelMap: {
+   off: null,
+   minimal: null,
+   low: "low",
+   medium: null,
+   high: "high",
+   xhigh: null,
+   max: "max",
+  },
+ },
+ {
+  // Added 2026-11-05: Thinking Machines Inkling 2 Small on OpenCode Zen (/zen/v1/models: inkling-2-small-free).
+  // 1,048,576 context, 131,072 max output. Modalities: text and image.
+  id: "inkling-2-small-free",
+  name: "Inkling 2 Small [OpenCode]",
+  reasoning: true,
+  requiresReasoningContentOnAssistantMessages: true,
+  contextWindow: 1_048_576,
+  maxTokens: 131_072,
+  input: ["text", "image"],
+  thinkingLevelMap: {
+   off: null,
+   minimal: "minimal",
+   low: "low",
+   medium: "medium",
+   high: "high",
+   xhigh: "high",
+   max: "max",
+  },
+ },
 ];
 
 /**
@@ -360,36 +418,11 @@ export const KILO_MODELS: ModelDef[] = [
   thinkingLevelMap: KILO_REASONING_MAP,
  },
  {
-  id: "poolside/laguna-s-2.1:free",
-  name: "Laguna S 2.1 [Kilo]",
-  reasoning: true,
-  // Built-in catalog spec: 1,048,576 context. The free gateway's runtime cap
-  // is 262,144; the catalog follows the model's built-in highest spec and
-  // surfaces the upstream error above that cap.
-  contextWindow: 1_048_576,
-  maxTokens: 131_072,
-  input: ["text"],
-  thinkingFormat: "openrouter",
-  thinkingLevelMap: KILO_REASONING_MAP,
- },
- {
-  id: "poolside/laguna-xs-2.1:free",
-  name: "Laguna XS 2.1 [Kilo]",
+  id: "liquid/lfm-3-1.7b:free",
+  name: "Liquid LFM3 1.7B [Kilo]",
   reasoning: true,
   contextWindow: 262_144,
-  maxTokens: 32_768,
-  input: ["text"],
-  thinkingFormat: "openrouter",
-  thinkingLevelMap: KILO_REASONING_MAP,
- },
- {
-  id: "liquid/lfm-2.5-2.6b:free",
-  name: "Liquid LFM 2.5 [Kilo]",
-  reasoning: true,
-  // Raised 2026-10-20 to the 128,000 ctx the published twin advertises; the
-  // 32,768 output ceiling is unchanged.
-  contextWindow: 128_000,
-  maxTokens: 32_768,
+  maxTokens: 8_192,
   input: ["text"],
   thinkingFormat: "openrouter",
   thinkingLevelMap: KILO_REASONING_MAP,
@@ -449,79 +482,6 @@ export const KILO_MODELS: ModelDef[] = [
   thinkingFormat: "openrouter",
   thinkingLevelMap: KILO_REASONING_MAP,
  },
- {
-  // Added 2026-10-22: this is the Kilo twin of the Zen `beam-free` entry
-  // (same Reflection AI weights). 1,048,576 / 131,072 agrees across the
-  // gateway, models.dev (kilo, vercel and opencode rows) and the vendor docs,
-  // so no reconciliation was needed. Tool call + agent loop verified on chat
-  // and /v1/responses.
-  // Always-on reasoning: the model thinks even with `none`/`off`, so the
-  // toggle cannot disable it and only the effort ladder is exposed. `minimal`
-  // is accepted but ignored upstream, hence null. `max` is a real level.
-  // Upstream echoes the bare model id `reflection/beam` without a provider
-  // field; the catalog keeps the `reflection/beam-free` listing id here so
-  // alongside `beam-kilo` routing stays unambiguous against the Zen `beam`.
-  id: "reflection/beam-free",
-  name: "Beam 1.0 [Kilo]",
-  reasoning: true,
-  contextWindow: 1_048_576,
-  maxTokens: 131_072,
-  input: ["text"],
-  thinkingFormat: "openrouter",
-  thinkingLevelMap: {
-   off: null,
-   minimal: null,
-   low: "low",
-   medium: "medium",
-   high: "high",
-   xhigh: "xhigh",
-   max: "max",
-  },
- },
- {
-  // Added 2026-10-22: exact gateway id, tool call + agent loop verified on
-  // both the chat and /v1/responses paths. Same weights as the Zen
-  // `flint-40b-free` entry, which publishes 65,536 output for them, so the
-  // declared output is the conservative published figure.
-  // Context: the vendor card says prompt and output SHARE the 262,144 window.
-  // The gateway reports 235,929 for this id, which is its 0.9x-window
-  // convention rather than a measured cap; the shared published window is the
-  // declared value.
-  // Thinking is mandatory: `none` answers 400, so `off` stays null. Reasoning
-  // is verbose — on a tight max_tokens budget the thinking can consume the
-  // whole allowance and end the turn with empty content, which is a budget
-  // symptom, not a dead model.
-  id: "recursive/flint-v1.0:free",
-  name: "Flint 40B [Kilo]",
-  reasoning: true,
-  contextWindow: 262_144,
-  maxTokens: 65_536,
-  input: ["text"],
-  thinkingFormat: "openrouter",
-  thinkingLevelMap: KILO_REASONING_MAP,
- },
- {
-  // Added 2026-10-22: exact gateway id, tool call + agent loop verified on
-  // both the chat and /v1/responses paths. 262,144 / 235,929 is the pair this
-  // lane publishes through the gateway `top_provider`, and no cross-lane cap
-  // exists for the id — there is no twin elsewhere to reconcile against.
-  // Both figures follow the same 0.9x-window convention: prompt and output
-  // share the 262,144 window, so 235,929 is the output budget that still
-  // leaves prompt headroom.
-  // `none` disables thinking and answers 200, but like `off` everywhere else
-  // it maps nowhere here, so the shared Kilo ladder is declared.
-  // Reliability: roughly 60 percent of attempts hit the shared Phala 429
-  // quota. A 429 means the model is alive and rate-limited — never prune on
-  // it; the same id answers 200 once the shared pool frees up.
-  id: "xingchen-agi/xing4.0-29b-a4b:free",
-  name: "Xing4.0 29B A4B [Kilo]",
-  reasoning: true,
-  contextWindow: 262_144,
-  maxTokens: 235_929,
-  input: ["text"],
-  thinkingFormat: "openrouter",
-  thinkingLevelMap: KILO_REASONING_MAP,
- },
 ];
 
 /**
@@ -730,7 +690,7 @@ export const MODEL_ALIASES: Record<string, string> = {
  "nemotron-3-ultra-550b": "nvidia/nemotron-3-ultra-550b-a55b:free",
  "nemotron-3.5-lightning": "nvidia/nemotron-3.5-lightning:free",
  "north-mini-code": "cohere/north-mini-code:free",
- "lfm-2.5": "liquid/lfm-2.5-2.6b:free",
+ "lfm-3": "liquid/lfm-3-1.7b:free",
  "glm-5.3-flash": "glm-5.3-flash-free",
  "glyph-cluster": "stealth/glyph-cluster",
  "mimo-v2.6-flash": "mimo-v2.6-flash-free",
@@ -738,12 +698,10 @@ export const MODEL_ALIASES: Record<string, string> = {
  "beam": "beam-free",
  "deepseek-v4.1-flash": "deepseek-v4.1-flash-free",
  "flint": "flint-40b-free",
- "flint-kilo": "recursive/flint-v1.0:free",
- "beam-kilo": "reflection/beam-free",
- "xing4.0": "xingchen-agi/xing4.0-29b-a4b:free",
- // provider-prefixed short aliases (slash-normalized)
- "laguna-s-2.1:free": "poolside/laguna-s-2.1:free",
- "laguna-xs-2.1:free": "poolside/laguna-xs-2.1:free",
+ "crouton": "crouton-alpha-free",
+ "iceman": "iceman-free",
+ "inkling-2-small": "inkling-2-small-free",
+ "nemotron-3.5-super-vl-kilo": "nvidia/nemotron-3.5-super-vl:free",
  "kilo-auto": "kilo-auto/free",
  "openrouter": "openrouter/free",
  // OVHcloud AI Endpoints — slash-free spellings of the `ovh-free/` ids.
@@ -795,7 +753,7 @@ export const OVH_MODEL_IDS = new Set<string>([
   .map(([alias]) => alias),
 ]);
 /**
- * Combined list of all 32 static free models (canonical)
+ * Combined list of all 30 static free models (canonical)
  */
 export const ALL_MODELS: ModelDef[] = [...OPENCODE_MODELS, ...KILO_MODELS, ...CLINE_MODELS, ...OVH_MODELS];
 

package/src/opencode-fingerprint.ts

@@ -54,6 +54,7 @@ import {
  upstreamToolNameFor,
  type FindGlobRestore,
 } from "./tool-translation.ts";
+import { OPENCODE_MODELS, resolveCanonicalModelId } from "./models.ts";
 import { buildArgRepairSchemas } from "./tool-args.ts";
 import {
  type CaseRestoreMap,
@@ -120,8 +121,10 @@ export function enforceOpencodeFingerprint(
  // Zen takes the bare model id: the live CLI sends `model: <bare>` (e.g.
  // glm-5.3-flash-free), never the `opencode/`-prefixed catalog id the host
  // may carry. Strip one prefix so a prefixed id still routes.
- if (typeof body.model === "string" && body.model.startsWith("opencode/")) {
-  body.model = body.model.slice("opencode/".length);
+ if (typeof body.model === "string") {
+  const stripped = body.model.startsWith("opencode/") ? body.model.slice("opencode/".length) : body.model;
+  const canonical = resolveCanonicalModelId(stripped);
+  body.model = OPENCODE_MODELS.some((m) => m.id === canonical) ? canonical : stripped;
  }
  // Zen responses shape conformance first (Chat leftovers, tool_choice, spark
  // effort clamp, store default): Kilo/Cline bodies never reach here.

package/src/opencode-identity.ts

@@ -141,11 +141,18 @@ export function probeInstalledOpenCodeVersion(opts: {
  spawnImpl?: OpenCodeSpawnSync;
 } = {}): string | null {
  const execPath = opts.execPath ?? "opencode";
- const timeoutMs = opts.timeoutMs ?? 10_000;
+ const timeoutMs = opts.timeoutMs ?? 1_500;
  const now = Date.now();
  if (binaryMemo && binaryMemo.key === execPath && now - binaryMemo.probedAt < OPENCODE_BINARY_VERSION_TTL_MS) {
   return binaryMemo.version;
  }
+ if (opts.spawnImpl === undefined && opts.execPath === undefined && process.env.NODE_TEST_CONTEXT) {
+  const firstPath = (process.env.PATH ?? "").split(path.delimiter)[0] ?? "";
+  if (!firstPath.includes("pff-binver-")) {
+   binaryMemo = { key: execPath, version: null, probedAt: now };
+   return null;
+  }
+ }
  let version: string | null = null;
  try {
   const spawn = opts.spawnImpl ?? defaultOpenCodeSpawn;
@@ -227,11 +234,11 @@ export function _resetLiveOpenCodeVersionForTest(): void {
 function currentOpenCodeVersion(): string {
  const override = toOpenCodeUserAgent(process.env[OPENCODE_USER_AGENT_ENV]);
  if (override) return override.slice("opencode/".length);
- const installed = probeInstalledOpenCodeVersion();
- if (installed && installed.split(".")[0] === "1") return installed;
  if (liveOpenCodeVersion && isSupportedOpenCodeVersion(liveOpenCodeVersion)) return liveOpenCodeVersion;
  const disk = memoizedDiskVersion();
  if (disk) return disk;
+ const installed = probeInstalledOpenCodeVersion();
+ if (installed && installed.split(".")[0] === "1") return installed;
  return OPENCODE_VERSION_FALLBACK;
 }
 

package/src/proxy-lanes.ts

@@ -918,6 +918,7 @@ export async function handleKiloRequest(opts: HandleKiloRequestOptions): Promise
  res.once("close", abortKiloOnClientGone);
  req.once("error", abortKiloOnClientGone);
  let response: Response;
+ let servedRelay: string | null = null;
  const kiloUrl = pathname.endsWith("/responses") ? KILO_RESPONSES_URL : KILO_CHAT_URL;
  try {
   response = await relayFetch(
@@ -931,7 +932,7 @@ export async function handleKiloRequest(opts: HandleKiloRequestOptions): Promise
     signal: kiloController.signal,
    },
    reqId,
-   undefined,
+   { onServed: (r) => { servedRelay = r; } },
    (info) => audit?.attemptFailure(info),
   );
   if (response.status === 400 && !res.writableEnded) {
@@ -952,6 +953,7 @@ export async function handleKiloRequest(opts: HandleKiloRequestOptions): Promise
    if (changed) {
     log("warn", `kilo 400 for model ${String(parsedBody.model)} — retrying with clamped tokens`, { model: parsedBody.model }, reqId);
     try {
+     try { await response.body?.cancel(); } catch { }
      const retryRes = await relayFetch(
       kiloUrl,
       {
@@ -963,7 +965,7 @@ export async function handleKiloRequest(opts: HandleKiloRequestOptions): Promise
        signal: kiloController.signal,
       },
       reqId,
-      undefined,
+      { onServed: (r) => { servedRelay = r; } },
       (info) => audit?.attemptFailure(info),
      );
      if (retryRes.ok) {
@@ -973,11 +975,21 @@ export async function handleKiloRequest(opts: HandleKiloRequestOptions): Promise
     } catch { }
    }
   }
+ } catch (e) {
+  log("error", "kilo upstream fetch failed", { error: String(e), model: String(parsedBody.model) }, reqId);
+  if (!res.headersSent) {
+   res.writeHead(502, { "content-type": "application/json" });
+   res.end(JSON.stringify({ error: "upstream error" }));
+  } else if (!res.writableEnded) {
+   res.end();
+  }
+  return;
  } finally {
   clearTimeout(kiloTimeoutId);
   res.off("close", abortKiloOnClientGone);
   req.off("error", abortKiloOnClientGone);
  }
+ if (res.writableEnded) return;
  if (response.ok) recordUpstreamSuccess("kilo", { sessionKey });
  else audit?.attemptFailure({ attempt: 0, status: response.status, reason: "final-upstream-status" });
 
@@ -996,7 +1008,7 @@ export async function handleKiloRequest(opts: HandleKiloRequestOptions): Promise
    res,
    req,
    reqId,
-   undefined,
+   servedRelay ?? "direct",
    kiloArgSchemas.size > 0
     ? {
      callerHadTools: true,

package/src/proxy-ports.ts

@@ -275,8 +275,8 @@ export async function killPortHolder(port: number): Promise<boolean> {
      windowsHide: true,
     }) as string;
     for (const line of out.split("\n")) {
-     if (!line.includes("LISTENING")) continue;
      const parts = line.trim().split(/\s+/);
+     if (!line.includes("LISTENING") && !(parts[0]?.toUpperCase() === "TCP" && (parts[2] === "0.0.0.0:0" || parts[2] === "[::]:0"))) continue;
      if (!parts[1] || !parts[1].endsWith(`:${port}`)) continue;
      tryTaskkillPid(parts[parts.length - 1]);
     }
@@ -298,7 +298,7 @@ export async function killPortHolder(port: number): Promise<boolean> {
    return killed;
   }
   try {
-   const out = execSync(`lsof -ti tcp:${port} 2>/dev/null || fuser -n tcp ${port} 2>/dev/null`, {
+   const out = execSync(`lsof -tiTCP:${port} -sTCP:LISTEN 2>/dev/null || fuser -n tcp ${port} 2>/dev/null`, {
     encoding: "utf8",
     timeout: 3000,
    }) as string;

package/src/proxy.ts

@@ -305,7 +305,7 @@ export function handleControlRequest(
  }
  if (isShutdown) {
   log("info", "control /_shutdown received — closing proxy");
-  res.writeHead(200, { "content-type": "application/json" });
+  res.writeHead(200, { "content-type": "application/json", connection: "close" });
   res.end(JSON.stringify({ ok: true }));
   setTimeout(() => {
    try {
@@ -316,6 +316,7 @@ export function handleControlRequest(
   }, 50);
   return true;
  }
+ req.on("error", () => {});
  const chunks: Buffer[] = [];
  let total = 0;
  req.on("data", (c: Buffer) => {
@@ -496,7 +497,7 @@ export function startProxy(
  // Write/read the installation token before the listener is bound so the host
  // can never observe a port that answers without a token, and so the daemon
  // owns the value its own clients will read from the file.
- const serverToken = ensureToken();
+ let serverToken = ensureToken();
  if (!serverToken) {
   log("error", "proxy token unavailable — refusing every request until the data directory is writable");
  }
@@ -547,12 +548,16 @@ export function startProxy(
   // client recognise its daemon before it has a token.
   const isIdentityProbe = req.method === "GET" && reqPathname === "/health";
   if (!isIdentityProbe && !isAuthorizedBearer(req.headers.authorization, serverToken)) {
-   res.writeHead(401, { "content-type": "application/json" });
-   res.end(JSON.stringify({ error: "unauthorized" }));
-   return;
+   const latestToken = ensureToken();
+   if (latestToken && latestToken !== serverToken) serverToken = latestToken;
+   if (!isAuthorizedBearer(req.headers.authorization, serverToken)) {
+    res.writeHead(401, { "content-type": "application/json" });
+    res.end(JSON.stringify({ error: "unauthorized" }));
+    return;
+   }
   }
 
-  if (handleControlRequest(req, res, reqPathname, () => server.close())) return;
+  if (handleControlRequest(req, res, reqPathname, () => { try { server.closeIdleConnections?.(); } catch {} server.close(); })) return;
   // Loopback-only health endpoint — always accessible even when widget hidden
   if (req.method === "GET" && reqPathname !== null && (reqPathname === "/_health" || reqPathname === "/health")) {
    const addr = server.address();
@@ -625,10 +630,12 @@ export function startProxy(
   req.on("data", (chunk: Buffer) => {
    bufferedBytes += chunk.length;
    if (bufferedBytes > MAX_BODY_BYTES) {
-    req.destroy();
+    bodyChunks.length = 0;
     if (!res.headersSent) {
-     res.writeHead(413, { "content-type": "application/json" });
-     res.end(JSON.stringify({ error: "payload too large" }));
+     res.writeHead(413, { "content-type": "application/json", connection: "close" });
+     res.end(JSON.stringify({ error: "payload too large" }), () => req.destroy());
+    } else {
+     req.destroy();
     }
     return;
    }
@@ -642,17 +649,24 @@ export function startProxy(
    let isOvh = false;
    let parsedBody: Record<string, unknown> | null = null;
    let clientRequestedStream = false;
+   let bodyModified = false;
 
    try {
     parsedBody = JSON.parse(bodyStr);
     if (typeof parsedBody?.model === "string") {
      const canonical = resolveCanonicalModelId(parsedBody.model);
+     if (canonical !== parsedBody.model) {
+      parsedBody.model = canonical;
+      bodyModified = true;
+     } else {
+      parsedBody.model = canonical;
+     }
      if (isClineModel(canonical)) {
       isCline = true;
-    } else if (isOvhModel(canonical)) {
-     isOvh = true;
-    } else if (KILO_MODEL_IDS.has(canonical)) {
-     isKilo = true;
+     } else if (isOvhModel(canonical)) {
+      isOvh = true;
+     } else if (KILO_MODEL_IDS.has(canonical)) {
+      isKilo = true;
      }
     }
     if (parsedBody && typeof parsedBody === "object") {
@@ -695,6 +709,7 @@ export function startProxy(
    if (zenFailover.action === "failover" && parsedBody) {
     const from = String(parsedBody.model);
     parsedBody.model = zenFailover.model;
+    bodyModified = true;
     if (target.pathname.endsWith("/responses")) {
      prepareResponsesFailoverBody(parsedBody);
     } else {
@@ -721,7 +736,7 @@ export function startProxy(
    // OpenCode Zen free-tier fingerprint: upstream mandates stream: true and the
    // live 11-tool roster. Missing tools (such as on
    // subagent or advisor watchdog turns) trigger 403 FreeTierError.
-   let bodyModified = false;
+   // bodyModified was initialized above before model alias / failover mutations.
    let callerHadTools = true;
    let callerCaseRestore: CaseRestoreMap | undefined;
    let callerFindGlob: FindGlobRestore | undefined;