fix(ai-chat): OpenAI Chat Completions for multi-turn + provider settings, stream UX & errors" -m "Live-stand fixes (OpenRouter / OpenAI-compatible):

- openai provider: use .chat() (Chat Completions) instead of the default callable
  (Responses API), which gateways reject on multi-turn -> 400.
- updateAiProviderSettings: assemble settings.ai.provider via jsonb_build_object
  with ::text-cast bound params + jsonb_typeof self-heal (postgres.js was
  double-encoding it into an array; the ::text cast avoids 'could not determine
  data type of parameter').
- chat agent: drop the hard maxOutputTokens cap (truncated complex tool calls);
  keep a tiny cap only on the test-connection ping.
- testConnection + chat stream: surface the real provider error (statusCode+message)
  to logs and the UI instead of generic masks; never log the API key.
- chat UI: typing indicator, incremental streaming render, tool 'running' status, Stop.

Also bundled (prior uncommitted ai-chat work):
- history 'AI agent' provenance badge; vector RAG (pgvector image + page_embeddings
  + AI_QUEUE indexer + space-scoped semanticSearch); external MCP servers backend
  (@ai-sdk/mcp client, SSRF IP-pinning, encrypted headers, admin CRUD/Test);
  yjs duplicate-instance fix via pnpm patch (single CJS instance server-side).

apps/server/src/integrations/ai/ai-embedding-not-configured.exception.ts

@@ -0,0 +1,13 @@
+import { ServiceUnavailableException } from '@nestjs/common';
+
+/**
+ * Thrown when no usable embedding config exists for the workspace (missing
+ * driver / embedding model / API key). Distinct from the chat variant so RAG
+ * callers (indexer / semanticSearch) can 503 or skip independently of chat
+ * being configured (§6.2/§6.7).
+ */
+export class AiEmbeddingNotConfiguredException extends ServiceUnavailableException {
+  constructor() {
+    super('AI embedding model not configured');
+  }
+}

apps/server/src/integrations/ai/ai.service.ts

@@ -1,10 +1,16 @@
-import { Injectable } from '@nestjs/common';
-import { generateText, type LanguageModel } from 'ai';
+import { Injectable, Logger } from '@nestjs/common';
+import {
+  embedMany,
+  generateText,
+  type EmbeddingModel,
+  type LanguageModel,
+} from 'ai';
 import { createOpenAI } from '@ai-sdk/openai';
 import { createGoogleGenerativeAI } from '@ai-sdk/google';
 import { createOllama } from 'ai-sdk-ollama';
 import { AiSettingsService } from './ai-settings.service';
 import { AiNotConfiguredException } from './ai-not-configured.exception';
+import { AiEmbeddingNotConfiguredException } from './ai-embedding-not-configured.exception';
 
 /**
  * Builds AI SDK language models from per-workspace config and runs cheap
@@ -16,6 +22,8 @@ import { AiNotConfiguredException } from './ai-not-configured.exception';
  */
 @Injectable()
 export class AiService {
+  private readonly logger = new Logger(AiService.name);
+
   constructor(private readonly aiSettings: AiSettingsService) {}
 
   /**
@@ -34,8 +42,13 @@ export class AiService {
 
     switch (cfg.driver) {
       case 'openai':
-        // baseURL (when set) covers openai-compatible endpoints.
-        return createOpenAI({ apiKey: cfg.apiKey, baseURL: cfg.baseUrl })(
+        // baseURL (when set) covers openai-compatible endpoints. Use Chat
+        // Completions (/chat/completions) — the portable OpenAI-compatible
+        // endpoint. The default callable createOpenAI(...)(model) targets the
+        // Responses API (/responses), which OpenAI-compatible gateways
+        // (OpenRouter, etc.) reject on multi-turn requests (history with
+        // assistant messages) → 400.
+        return createOpenAI({ apiKey: cfg.apiKey, baseURL: cfg.baseUrl }).chat(
           cfg.chatModel,
         );
       case 'gemini':
@@ -48,34 +61,90 @@ export class AiService {
     }
   }
 
+  /**
+   * Resolve the workspace config and build the text-embedding model used by the
+   * RAG indexer / semanticSearch (§6.7 stage D). Built PER WORKSPACE on demand,
+   * same as getChatModel; the decrypted key is never logged.
+   *
+   * Throws AiEmbeddingNotConfiguredException (→ 503) when the driver,
+   * embeddingModel or (for non-ollama) the API key is missing, so RAG callers
+   * can 503 or skip independently of chat being configured.
+   */
+  async getEmbeddingModel(workspaceId: string): Promise<EmbeddingModel> {
+    const cfg = await this.aiSettings.resolve(workspaceId);
+    if (
+      !cfg?.driver ||
+      !cfg?.embeddingModel ||
+      (cfg.driver !== 'ollama' && !cfg.apiKey)
+    ) {
+      throw new AiEmbeddingNotConfiguredException();
+    }
+
+    switch (cfg.driver) {
+      case 'openai':
+        // baseURL (when set) covers openai-compatible endpoints.
+        return createOpenAI({
+          apiKey: cfg.apiKey,
+          baseURL: cfg.baseUrl,
+        }).textEmbeddingModel(cfg.embeddingModel);
+      case 'gemini':
+        return createGoogleGenerativeAI({
+          apiKey: cfg.apiKey,
+        }).textEmbeddingModel(cfg.embeddingModel);
+      case 'ollama':
+        // Ollama needs no API key (e.g. nomic-embed-text).
+        return createOllama({ baseURL: cfg.baseUrl }).textEmbeddingModel(
+          cfg.embeddingModel,
+        );
+      default:
+        throw new AiEmbeddingNotConfiguredException();
+    }
+  }
+
+  /**
+   * Embed a batch of texts with the workspace embedding model. Returns one
+   * vector per input, in the same order. Thin wrapper over the AI SDK's
+   * embedMany; never logs the key or the texts.
+   */
+  async embedTexts(workspaceId: string, texts: string[]): Promise<number[][]> {
+    if (texts.length === 0) return [];
+    const model = await this.getEmbeddingModel(workspaceId);
+    const { embeddings } = await embedMany({ model, values: texts });
+    return embeddings;
+  }
+
   /**
    * Cheap connectivity check. Builds the model and asks for a one-word reply.
-   * Never leaks the provider's raw error body or the key — only a short,
-   * generic message (§6.4/§8.3).
+   * On AiNotConfiguredException returns a generic "not configured" message; for
+   * any other failure surfaces the provider's own cause (e.g. AI SDK
+   * `AI_APICallError` -> `${statusCode}: ${message}`) so a 402 / wrong model /
+   * missing key is diagnosable, and logs the full error. The decrypted key is
+   * never logged or returned — AI SDK error messages/4xx bodies do not contain
+   * it, and the resolved config (which holds the key) is never dumped (§6.4/§8.3).
    */
   async testConnection(
     workspaceId: string,
   ): Promise<{ ok: true } | { ok: false; error: string }> {
-    let model: LanguageModel;
     try {
-      model = await this.getChatModel(workspaceId);
+      const model = await this.getChatModel(workspaceId);
+      // maxOutputTokens keeps the probe cheap and avoids providers (e.g.
+      // OpenRouter) reserving/charging for the model's full max-token budget,
+      // which would 402 on a key with limited credit.
+      await generateText({ model, prompt: 'ping', maxOutputTokens: 16 });
+      return { ok: true };
     } catch (err) {
       if (err instanceof AiNotConfiguredException) {
         return { ok: false, error: 'AI provider not configured' };
       }
-      // Defensive: do not surface internal error details.
-      return { ok: false, error: 'AI provider not configured' };
-    }
-
-    try {
-      await generateText({ model, prompt: 'ping' });
-      return { ok: true };
-    } catch {
-      // Do NOT include the provider's raw error (may echo the request/key).
-      return {
-        ok: false,
-        error: 'Failed to reach the AI provider. Check the settings and key.',
-      };
+      // Surface the real provider cause so failures are diagnosable, and log the
+      // full error. AI SDK errors expose statusCode/message (and responseBody);
+      // none of these carry the key. Do NOT log/return the resolved config.
+      this.logger.error('AI test connection failed', err as Error);
+      const e = err as { statusCode?: number; message?: string };
+      const msg = e?.statusCode
+        ? `${e.statusCode}: ${e.message}`
+        : (e?.message ?? 'Unknown error');
+      return { ok: false, error: msg };
     }
   }
 }

apps/server/src/integrations/queue/constants/queue.interface.ts

@@ -22,6 +22,17 @@ export interface IPageHistoryJob {
   pageId: string;
 }
 
+/**
+ * AI_QUEUE payload for a content change that should trigger a RAG reindex
+ * (§6.7 stage D / §14[M1]). Produced by the collab persistence extension on
+ * `onStoreDocument` and by the page-delete path (the delete case carries the
+ * ids of pages whose embeddings must be purged).
+ */
+export interface IPageContentUpdatedJob {
+  pageIds: string[];
+  workspaceId: string;
+}
+
 export interface INotificationCreateJob {
   userId: string;
   workspaceId: string;