@selesai/code 0.13.11 → 0.13.12

package/CHANGELOG.md

@@ -2,6 +2,11 @@
 
 All notable changes to `@selesai/code` will be documented in this file.
 
+## [0.13.12] - 2026-09-03
+
+### Added
+- **Auto-continue after output token limit.** When the model stops because it reached the maximum output token limit (`stopReason: "length"`) mid-response, the agent now queues a follow-up "continue from where you left off" message and re-prompts automatically instead of stopping with the "response may be incomplete" error. Truncated tool calls were already re-issued by the agent loop; this covers truncated text. The continuation is capped at 3 consecutive length stops (the counter resets on any normal stop) so a model that keeps hitting the cap doesn't loop forever, and context-pressure length stops still trigger compaction first.
+
 ## [0.13.11] - 2026-09-02
 
 ### Added

package/dist/core/agent-session-length-continuation.test.d.ts

@@ -0,0 +1 @@
+export {};

package/dist/core/agent-session-length-continuation.test.js

@@ -0,0 +1,107 @@
+import { describe, expect, it, vi } from "vitest";
+import { AgentSession } from "./agent-session.js";
+import { SettingsManager } from "./settings-manager.js";
+function createMockSession() {
+    const settings = SettingsManager.inMemory({});
+    const extensionRunner = {
+        getCommand: vi.fn(),
+        createCommandContext: vi.fn(),
+        emitError: vi.fn(),
+        emit: vi.fn(),
+        emitMessageEnd: vi.fn(),
+        hasHandlers: vi.fn(() => false),
+    };
+    const resourceLoader = {
+        getExtensions: () => ({ extensions: [], errors: [], runtime: { flagValues: new Map(), pendingProviderRegistrations: [], pendingNativeProviderRegistrations: [], assertActive: () => { } } }),
+        getSkills: () => ({ skills: [], diagnostics: [] }),
+        getPrompts: () => ({ prompts: [], diagnostics: [] }),
+        getThemes: () => ({ themes: [], diagnostics: [] }),
+        getAgents: () => ({ agents: [], diagnostics: [] }),
+        getAgentsFiles: () => ({ agentsFiles: [] }),
+        getSystemPrompt: () => undefined,
+        getAppendSystemPrompt: () => [],
+        getExtensionDiagnostics: () => [],
+        extendResources: () => { },
+        reload: vi.fn(),
+    };
+    const modelRegistry = {
+        registerProvider: vi.fn(),
+        unregisterProvider: vi.fn(),
+        getProvider: vi.fn(),
+        getProviders: vi.fn(() => []),
+        getModels: vi.fn(() => []),
+    };
+    const followUp = vi.fn();
+    const hasQueuedMessages = vi.fn(() => false);
+    const agent = {
+        state: { messages: [], tools: [] },
+        subscribe: vi.fn(() => () => { }),
+        followUp,
+        hasQueuedMessages,
+    };
+    const session = new AgentSession({
+        agent: agent,
+        sessionManager: { getBranch: () => [], appendMessage: vi.fn() },
+        settingsManager: settings,
+        cwd: "/tmp",
+        resourceLoader: resourceLoader,
+        modelRuntime: modelRegistry,
+    });
+    session._extensionRunner = extensionRunner;
+    return { session, followUp, hasQueuedMessages };
+}
+function lengthStopMessage(text) {
+    return {
+        role: "assistant",
+        content: [{ type: "text", text }],
+        stopReason: "length",
+        timestamp: Date.now(),
+        usage: { input: 100, output: 500, cacheRead: 0, cacheWrite: 0, totalTokens: 600, cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 } },
+    };
+}
+describe("AgentSession length-stop continuation", () => {
+    it("queues a follow-up continuation and returns true for a truncated text response", async () => {
+        const { session, followUp } = createMockSession();
+        session._lastAssistantMessage = lengthStopMessage("partial response...");
+        const result = await session._handlePostAgentRun();
+        expect(result).toBe(true);
+        expect(followUp).toHaveBeenCalledTimes(1);
+        const queued = followUp.mock.calls[0][0];
+        expect(queued.role).toBe("user");
+        expect(queued.content[0].text).toContain("Continue from where you left off");
+    });
+    it("does not continue when the truncated message contains tool calls", async () => {
+        const { session, followUp } = createMockSession();
+        session._lastAssistantMessage = {
+            role: "assistant",
+            content: [{ type: "toolCall", id: "tc_1", name: "bash", arguments: "{}" }],
+            stopReason: "length",
+            timestamp: Date.now(),
+        };
+        const result = await session._handlePostAgentRun();
+        expect(result).toBe(false);
+        expect(followUp).not.toHaveBeenCalled();
+    });
+    it("stops after MAX_LENGTH_CONTINUATIONS consecutive length stops", async () => {
+        const { session, followUp } = createMockSession();
+        session._lengthContinuations = 3;
+        session._lastAssistantMessage = lengthStopMessage("still truncated...");
+        const result = await session._handlePostAgentRun();
+        expect(result).toBe(false);
+        expect(followUp).not.toHaveBeenCalled();
+    });
+    it("resets the continuation counter on a non-length assistant stop", async () => {
+        const { session } = createMockSession();
+        session._lengthContinuations = 2;
+        await session._handleAgentEvent({
+            type: "message_end",
+            message: {
+                role: "assistant",
+                content: [{ type: "text", text: "done" }],
+                stopReason: "stop",
+                timestamp: Date.now(),
+            },
+        });
+        expect(session._lengthContinuations).toBe(0);
+    });
+});

package/dist/core/agent-session.d.ts

@@ -232,6 +232,7 @@ export declare class AgentSession {
     private _branchSummaryAbortController;
     private _retryAbortController;
     private _retryAttempt;
+    private _lengthContinuations;
     private readonly _bashAbortControllers;
     private _pendingBashMessages;
     private _extensionRunner;

package/dist/core/agent-session.js

@@ -82,6 +82,10 @@ function estimateMessagesTokens(messages) {
 // ============================================================================
 // Constants
 // ============================================================================
+/** Max consecutive auto-continuations after a length stop (output token limit). */
+const MAX_LENGTH_CONTINUATIONS = 3;
+/** Follow-up message queued when the model hits the output token limit. */
+const LENGTH_CONTINUATION_MESSAGE = "Your previous response was cut off because it reached the maximum output token limit. Continue from where you left off. Do not repeat content you already produced.";
 // ============================================================================
 // AgentSession Class
 // ============================================================================
@@ -113,6 +117,8 @@ export class AgentSession {
     // Retry state
     _retryAbortController = undefined;
     _retryAttempt = 0;
+    // Length-stop continuation state (consecutive output-token-limit continuations)
+    _lengthContinuations = 0;
     // Bash execution state
     _bashAbortControllers = new Set();
     _pendingBashMessages = [];
@@ -441,6 +447,11 @@ export class AgentSession {
                 if (assistantMsg.stopReason !== "error") {
                     this._overflowRecoveryAttempted = false;
                 }
+                // Reset the length-continuation counter on any non-length stop so the
+                // cap only bounds consecutive output-token-limit continuations.
+                if (assistantMsg.stopReason !== "length") {
+                    this._lengthContinuations = 0;
+                }
                 // Reset retry counter immediately on successful assistant response
                 // This prevents accumulation across multiple LLM calls within a turn
                 if (assistantMsg.stopReason !== "error" && this._retryAttempt > 0) {
@@ -863,6 +874,18 @@ export class AgentSession {
         if (await this._checkCompaction(msg)) {
             return true;
         }
+        // A length stop means the model hit the output token limit mid-response.
+        // The agent loop already re-prompts when truncated tool calls are present;
+        // here the loop stopped on truncated text, so queue a follow-up to continue.
+        // Compaction runs first: a length stop caused by context pressure (input
+        // filling the window) should compact instead of blindly continuing.
+        if (msg.stopReason === "length" &&
+            !msg.content.some((c) => c.type === "toolCall") &&
+            this._lengthContinuations < MAX_LENGTH_CONTINUATIONS) {
+            this._lengthContinuations++;
+            await this._queueFollowUp(LENGTH_CONTINUATION_MESSAGE);
+            return true;
+        }
         // The agent loop drains both queues before emitting agent_end. Any messages
         // here were queued by agent_end extension handlers and need a continuation.
         return this.agent.hasQueuedMessages();

package/dist/extensions/pi-hermes-memory/src/index.ts

@@ -24,7 +24,7 @@
 
 import * as path from "node:path";
 import * as fs from "node:fs";
-import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
+import type { ExtensionAPI, ExtensionContext } from "@earendil-works/pi-coding-agent";
 import { MemoryStore } from "./store/memory-store.js";
 import { SkillStore } from "./store/skill-store.js";
 import { DatabaseManager } from "./store/db.js";
@@ -106,6 +106,10 @@ export default function (pi: ExtensionAPI) {
 
   const shouldMigrateExtensionRoot = !configuredMemoryDir || pointsToLegacyMemoryDir;
   let persistenceInitialized = false;
+  // Captured at session_start: the extension factory runs with no session, but
+  // the auto-consolidator needs model access to use the fast in-process direct
+  // transport instead of always spawning a slow `pi -p` subprocess agent.
+  let sessionModelCtx: Pick<ExtensionContext, "model" | "modelRegistry"> | null = null;
 
   const store = new MemoryStore({ ...config, memoryDir: globalDir });
   // Factory may run with no session (Pi public contract). Do not snapshot
@@ -182,6 +186,7 @@ export default function (pi: ExtensionAPI) {
 
   // ── 1. Load memory from disk on session start ──
   pi.on("session_start", async (_event, ctx) => {
+    sessionModelCtx = { model: ctx.model, modelRegistry: ctx.modelRegistry };
     if (!persistenceInitialized) {
       try {
         await measureLifecycle("session-start.persistence-sync", async () => {
@@ -286,6 +291,7 @@ export default function (pi: ExtensionAPI) {
       config.consolidationTimeoutMs,
       toolTarget,
       config,
+      sessionModelCtx,
     );
     if (result.deferred) {
       console.info(`⏳ Auto-consolidation for '${toolTarget}' deferred: ${result.error ?? "another session holds the consolidation lock"}`);

package/dist/extensions/pi-web-agent/src/backends/config.ts

@@ -17,7 +17,7 @@ export type FirecrawlOptions = {
 };
 
 export type SearchBackendConfig = {
-  provider: 'duckduckgo' | 'searxng' | 'brave' | 'youcom' | 'exa' | 'tavily';
+  provider: 'duckduckgo' | 'searxng' | 'brave' | 'youcom' | 'exa' | 'tavily' | 'tokenin';
   baseUrl?: string;
   fallback?: 'duckduckgo';
   options?: SearxngOptions;
@@ -54,7 +54,7 @@ export type BackendConfigFile = {
 };
 
 export const DEFAULT_BACKEND_CONFIG: BackendConfig = {
-  search: { provider: 'duckduckgo' },
+  search: { provider: 'tokenin', fallback: 'duckduckgo' },
   fetch: { provider: 'http' },
   headless: { provider: 'local-browser' }
 };
@@ -86,7 +86,7 @@ function extractFirecrawlOptions(value: unknown): FirecrawlOptions | undefined {
   return Object.keys(options).length > 0 ? options : undefined;
 }
 
-const PROVIDER_NAMES: SearchProviderName[] = ['duckduckgo', 'searxng', 'brave', 'youcom', 'exa', 'tavily'];
+const PROVIDER_NAMES: SearchProviderName[] = ['duckduckgo', 'searxng', 'brave', 'youcom', 'exa', 'tavily', 'tokenin'];
 
 export function usableSearchProviders(
   search: SearchBackendConfig,
@@ -99,6 +99,7 @@ export function usableSearchProviders(
   if (env.YDC_API_KEY?.trim()) usable.push('youcom');
   if (env.EXA_API_KEY?.trim()) usable.push('exa');
   if (env.TAVILY_API_KEY?.trim()) usable.push('tavily');
+  if (hasActiveTokenInAccount()) usable.push('tokenin');
   return usable;
 }
 
@@ -129,7 +130,8 @@ export function extractBackendConfigOverride(
     backends?.search?.provider === 'brave' ||
     backends?.search?.provider === 'youcom' ||
     backends?.search?.provider === 'exa' ||
-    backends?.search?.provider === 'tavily'
+    backends?.search?.provider === 'tavily' ||
+    backends?.search?.provider === 'tokenin'
   ) {
     override.search = { provider: backends.search.provider };
     if (backends.search.provider === 'searxng' && typeof backends.search.baseUrl === 'string') {
@@ -186,8 +188,8 @@ export function validateBackendConfig(config: BackendConfig): string[] {
     issues.push('fetch provider firecrawl requires backends.fetch.baseUrl');
   }
 
-  if (config.search.fallback === 'duckduckgo' && config.search.provider !== 'searxng' && config.search.provider !== 'brave' && config.search.provider !== 'youcom' && config.search.provider !== 'exa' && config.search.provider !== 'tavily') {
-    issues.push('search fallback duckduckgo is only supported when search provider is searxng, brave, youcom, exa, or tavily');
+  if (config.search.fallback === 'duckduckgo' && config.search.provider !== 'searxng' && config.search.provider !== 'brave' && config.search.provider !== 'youcom' && config.search.provider !== 'exa' && config.search.provider !== 'tavily' && config.search.provider !== 'tokenin') {
+    issues.push('search fallback duckduckgo is only supported when search provider is searxng, brave, youcom, exa, tavily, or tokenin');
   }
 
   if (config.fetch.fallback === 'http' && config.fetch.provider !== 'firecrawl') {

package/dist/extensions/pi-web-agent/src/backends/factory.ts

@@ -3,6 +3,7 @@ import { createBraveSearchTool } from '../search/brave.js';
 import { createYouComSearchTool } from '../search/youcom.js';
 import { createExaSearchTool } from '../search/exa.js';
 import { createTavilySearchTool } from '../search/tavily.js';
+import { createTokenInSearchTool } from '../search/tokenin.js';
 import { createSearxngSearchTool } from '../search/searxng.js';
 import { createFanoutSearch } from '../search/fanout.js';
 import { buildFetchPresentation } from '../presentation/fetch-presentation.js';
@@ -31,6 +32,7 @@ export type BackendFactoryDeps = {
   createYouComSearch?: typeof createYouComSearchTool;
   createExaSearch?: typeof createExaSearchTool;
   createTavilySearch?: typeof createTavilySearchTool;
+  createTokenInSearch?: typeof createTokenInSearchTool;
   createHttpFetch?: typeof createWebFetchTool;
   createFirecrawlFetch?: typeof createFirecrawlFetcher;
   createHeadlessFetch?: typeof createWebFetchHeadlessTool;
@@ -71,7 +73,7 @@ function invalidFirecrawlFetch() {
 function withSearchFallback(
   primary: BackendSet['search'],
   fallback: BackendSet['search'],
-  fallbackFrom: 'searxng' | 'brave' | 'youcom' | 'exa' | 'tavily' | 'duckduckgo'
+  fallbackFrom: 'searxng' | 'brave' | 'youcom' | 'exa' | 'tavily' | 'tokenin' | 'duckduckgo'
 ): BackendSet['search'] {
   return async (input) => {
     const first = await primary(input);
@@ -124,6 +126,7 @@ export function createBackendSet(
   const createYouComSearch = deps.createYouComSearch ?? createYouComSearchTool;
   const createExaSearch = deps.createExaSearch ?? createExaSearchTool;
   const createTavilySearch = deps.createTavilySearch ?? createTavilySearchTool;
+  const createTokenInSearch = deps.createTokenInSearch ?? createTokenInSearchTool;
   const createHttpFetch = deps.createHttpFetch ?? createWebFetchTool;
   const createFirecrawlFetch = deps.createFirecrawlFetch ?? createFirecrawlFetcher;
   const createHeadlessFetch = deps.createHeadlessFetch ?? createWebFetchHeadlessTool;
@@ -142,6 +145,8 @@ export function createBackendSet(
         return createExaSearch({ apiKey: process.env.EXA_API_KEY });
       case 'tavily':
         return createTavilySearch({ apiKey: process.env.TAVILY_API_KEY });
+      case 'tokenin':
+        return createTokenInSearch();
       case 'duckduckgo':
       default:
         return createDuckDuckGoSearch();
@@ -160,7 +165,9 @@ export function createBackendSet(
           ? createExaSearch({ apiKey: process.env.EXA_API_KEY })
           : config.search.provider === 'tavily'
             ? createTavilySearch({ apiKey: process.env.TAVILY_API_KEY })
-            : createDuckDuckGoSearch();
+            : config.search.provider === 'tokenin'
+              ? createTokenInSearch()
+              : createDuckDuckGoSearch();
 
   if (config.search.provider === 'searxng' && config.search.fallback === 'duckduckgo') {
     search = withSearchFallback(search, createDuckDuckGoSearch(), 'searxng');
@@ -182,6 +189,10 @@ export function createBackendSet(
     search = withSearchFallback(search, createDuckDuckGoSearch(), 'tavily');
   }
 
+  if (config.search.provider === 'tokenin' && config.search.fallback === 'duckduckgo') {
+    search = withSearchFallback(search, createDuckDuckGoSearch(), 'tokenin');
+  }
+
   const fanoutConfig = config.search.fanout;
   if (fanoutConfig && fanoutConfig.mode !== 'off') {
     const baseNames =

package/dist/extensions/pi-web-agent/src/backends/settings-reader.ts

@@ -4,7 +4,28 @@
 // minimal settings.json read. Ceiling: if selesai moves web search config elsewhere, this
 // helper needs updating; upgrade path = upstream pi-web-agent adopting a settings hook.
 import { existsSync, readFileSync } from 'node:fs';
-import { getSettingsPath } from '@selesai/code';
+import { join } from 'node:path';
+import { getAgentDir, getSettingsPath } from '@selesai/code';
+
+/**
+ * True when tokenin-auth.json (written by the tokenin-onboarding extension)
+ * has an active account. Used to decide whether the tokenin search provider is
+ * usable. Never throws.
+ */
+export function hasActiveTokenInAccount(authPath: string = join(getAgentDir(), 'tokenin-auth.json')): boolean {
+	try {
+		if (!existsSync(authPath)) return false;
+		const parsed = JSON.parse(readFileSync(authPath, 'utf-8')) as unknown;
+		if (!parsed || typeof parsed !== 'object' || Array.isArray(parsed)) return false;
+		const auth = parsed as { accounts?: unknown; activeId?: unknown };
+		if (!Array.isArray(auth.accounts) || typeof auth.activeId !== 'string') return false;
+		return auth.accounts.some(
+			(a) => a && typeof a === 'object' && !Array.isArray(a) && (a as { id?: unknown }).id === auth.activeId
+		);
+	} catch {
+		return false;
+	}
+}
 
 /**
  * Read the Brave API key from selesai settings.json (webAgent.braveApiKey).

package/dist/extensions/pi-web-agent/src/commands/web-agent-config.ts

@@ -278,7 +278,7 @@ function buildBackendSettingsItems(
       id: 'backend:search:provider',
       label: 'Search backend',
       currentValue: backends.search.provider,
-      values: ['duckduckgo', 'searxng', 'brave', 'youcom', 'exa', 'tavily']
+      values: ['duckduckgo', 'searxng', 'brave', 'youcom', 'exa', 'tavily', 'tokenin']
     },
     {
       id: 'backend:search:baseUrl',
@@ -289,8 +289,8 @@ function buildBackendSettingsItems(
     {
       id: 'backend:search:fallback',
       label: 'Search fallback',
-      currentValue: backends.search.provider === 'searxng' || backends.search.provider === 'brave' || backends.search.provider === 'youcom' || backends.search.provider === 'exa' || backends.search.provider === 'tavily' ? backends.search.fallback ?? 'off' : 'off',
-      values: backends.search.provider === 'searxng' || backends.search.provider === 'brave' || backends.search.provider === 'youcom' || backends.search.provider === 'exa' || backends.search.provider === 'tavily' ? ['off', 'duckduckgo'] : ['off']
+      currentValue: backends.search.provider === 'searxng' || backends.search.provider === 'brave' || backends.search.provider === 'youcom' || backends.search.provider === 'exa' || backends.search.provider === 'tavily' || backends.search.provider === 'tokenin' ? backends.search.fallback ?? 'off' : 'off',
+      values: backends.search.provider === 'searxng' || backends.search.provider === 'brave' || backends.search.provider === 'youcom' || backends.search.provider === 'exa' || backends.search.provider === 'tavily' || backends.search.provider === 'tokenin' ? ['off', 'duckduckgo'] : ['off']
     },
     {
       id: 'backend:search:fanout:mode',
@@ -488,7 +488,7 @@ export function applySettingsValue(
     currentDraft.tools = nextTools;
   }
 
-  if (id === 'backend:search:provider' && (newValue === 'duckduckgo' || newValue === 'searxng' || newValue === 'brave' || newValue === 'youcom' || newValue === 'exa' || newValue === 'tavily')) {
+  if (id === 'backend:search:provider' && (newValue === 'duckduckgo' || newValue === 'searxng' || newValue === 'brave' || newValue === 'youcom' || newValue === 'exa' || newValue === 'tavily' || newValue === 'tokenin')) {
     currentBackends.search.provider = newValue;
     if (newValue !== 'searxng') {
       delete currentBackends.search.baseUrl;
@@ -500,7 +500,7 @@ export function applySettingsValue(
   }
 
   if (id === 'backend:search:fallback') {
-    if (newValue === 'duckduckgo' && (currentBackends.search.provider === 'searxng' || currentBackends.search.provider === 'brave' || currentBackends.search.provider === 'youcom' || currentBackends.search.provider === 'exa' || currentBackends.search.provider === 'tavily')) {
+    if (newValue === 'duckduckgo' && (currentBackends.search.provider === 'searxng' || currentBackends.search.provider === 'brave' || currentBackends.search.provider === 'youcom' || currentBackends.search.provider === 'exa' || currentBackends.search.provider === 'tavily' || currentBackends.search.provider === 'tokenin')) {
       currentBackends.search.fallback = 'duckduckgo';
     } else {
       delete currentBackends.search.fallback;

package/dist/extensions/pi-web-agent/src/search/tokenin.ts

@@ -0,0 +1,130 @@
+import { existsSync, readFileSync } from 'node:fs';
+import { join } from 'node:path';
+import { getAgentDir } from '@selesai/code';
+import { buildSearchPresentation } from '../presentation/search-presentation.js';
+import type { SearchResult, WebSearchResponse } from '../types.js';
+
+export const TOKENIN_SEARCH_TOOL_NAME = 'firecrawl';
+export const TOKENIN_DEFAULT_BASE_URL = 'https://lite.andlet.me/v1';
+
+type TokenInAccount = { id: string; label: string; apiKey: string; baseUrl?: string };
+type TokenInAuth = { accounts: TokenInAccount[]; activeId: string | null };
+
+type LiteLLMSearchResult = { title?: unknown; url?: unknown; snippet?: unknown };
+type LiteLLMSearchResponse = { results?: LiteLLMSearchResult[] };
+
+/**
+ * Read the active Token-In account from tokenin-auth.json (written by the
+ * tokenin-onboarding extension). Returns undefined when no account is saved.
+ * Never throws.
+ */
+export function readActiveTokenInAccount(
+  authPath: string = join(getAgentDir(), 'tokenin-auth.json')
+): TokenInAccount | undefined {
+  try {
+    if (!existsSync(authPath)) return undefined;
+    const parsed = JSON.parse(readFileSync(authPath, 'utf-8')) as unknown;
+    if (!parsed || typeof parsed !== 'object' || Array.isArray(parsed)) return undefined;
+    const auth = parsed as TokenInAuth;
+    if (!Array.isArray(auth.accounts) || typeof auth.activeId !== 'string') return undefined;
+    return auth.accounts.find((a) => a && typeof a === 'object' && a.id === auth.activeId && typeof a.apiKey === 'string');
+  } catch {
+    return undefined;
+  }
+}
+
+function resultWithPresentation(result: WebSearchResponse): WebSearchResponse {
+  return { ...result, presentation: buildSearchPresentation(result) };
+}
+
+function normalizeResults(response: LiteLLMSearchResponse): SearchResult[] {
+  return (response.results ?? []).flatMap((item) => {
+    if (typeof item.title !== 'string' || typeof item.url !== 'string') return [];
+    return [
+      {
+        title: item.title,
+        url: item.url,
+        snippet: typeof item.snippet === 'string' ? item.snippet : ''
+      }
+    ];
+  });
+}
+
+export function createTokenInSearchTool({
+  readAccount = readActiveTokenInAccount,
+  fetchImpl = fetch
+}: {
+  readAccount?: () => TokenInAccount | undefined;
+  fetchImpl?: typeof fetch;
+} = {}) {
+  return async function tokenInSearch({ query }: { query: string }): Promise<WebSearchResponse> {
+    const normalizedQuery = query.trim();
+
+    if (!normalizedQuery) {
+      return resultWithPresentation({
+        status: 'error',
+        results: [],
+        metadata: { backend: 'tokenin', cacheHit: false },
+        error: { code: 'INVALID_QUERY', message: 'Query must not be empty.' }
+      });
+    }
+
+    const account = readAccount();
+    if (!account) {
+      return resultWithPresentation({
+        status: 'error',
+        results: [],
+        metadata: { backend: 'tokenin', cacheHit: false },
+        error: {
+          code: 'BACKEND_CONFIG_INVALID',
+          message: 'Token-In search requires an active Token-In account. Run /tokenin add first.'
+        }
+      });
+    }
+
+    const baseUrl = (account.baseUrl ?? TOKENIN_DEFAULT_BASE_URL).replace(/\/+$/, '');
+    const url = `${baseUrl}/v1/search/${TOKENIN_SEARCH_TOOL_NAME}`;
+
+    try {
+      const response = await fetchImpl(url, {
+        method: 'POST',
+        headers: {
+          Accept: 'application/json',
+          'Content-Type': 'application/json',
+          Authorization: `Bearer ${account.apiKey}`
+        },
+        body: JSON.stringify({ query: normalizedQuery, max_results: 10 })
+      });
+
+      if (!response.ok) {
+        throw new Error(`HTTP ${response.status}`);
+      }
+
+      const parsed = (await response.json()) as LiteLLMSearchResponse;
+      const results = normalizeResults(parsed);
+
+      if (results.length === 0) {
+        return resultWithPresentation({
+          status: 'error',
+          results: [],
+          metadata: { backend: 'tokenin', cacheHit: false },
+          error: { code: 'NO_RESULTS', message: 'Token-In search returned no usable results for this query.' }
+        });
+      }
+
+      return resultWithPresentation({
+        status: 'ok',
+        results,
+        metadata: { backend: 'tokenin', cacheHit: false }
+      });
+    } catch (error) {
+      const rawMessage = error instanceof Error ? error.message : String(error);
+      return resultWithPresentation({
+        status: 'error',
+        results: [],
+        metadata: { backend: 'tokenin', cacheHit: false },
+        error: { code: 'FETCH_FAILED', message: `Token-In search request failed: ${rawMessage}` }
+      });
+    }
+  };
+}

package/dist/extensions/pi-web-agent/src/types.ts

@@ -16,7 +16,7 @@ export type SearchResult = {
   snippet: string;
 };
 
-export type SearchProviderName = 'duckduckgo' | 'searxng' | 'brave' | 'youcom' | 'exa' | 'tavily';
+export type SearchProviderName = 'duckduckgo' | 'searxng' | 'brave' | 'youcom' | 'exa' | 'tavily' | 'tokenin';
 
 export type FanoutMode = 'off' | 'on' | 'auto';
 
@@ -32,9 +32,9 @@ export type ToolError = {
 };
 
 export type SearchMetadata = {
-  backend: 'duckduckgo' | 'searxng' | 'brave' | 'youcom' | 'exa' | 'tavily';
+  backend: SearchProviderName;
   cacheHit: boolean;
-  fallbackFrom?: 'searxng' | 'brave' | 'youcom' | 'exa' | 'tavily' | 'duckduckgo';
+  fallbackFrom?: Exclude<SearchProviderName, 'tokenin'>;
   fallbackReason?: string;
   fanout?: FanoutMetadata;
 };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@selesai/code",
-  "version": "0.13.11",
+  "version": "0.13.12",
   "description": "Maintained, extension-first Pi coding agent with built-in workflows, subagents, web research, questions, skills, and an enhanced terminal UI.",
   "type": "module",
   "repository": {