brainjar 0.1.0 → 0.2.0

package/dist/brainjar.js

@@ -174,8 +174,8 @@ var init_github = __esm({
         if (this.opts.token) h.Authorization = `Bearer ${this.opts.token}`;
         return h;
       }
-      async *paginate(path6, params) {
-        let url = this.buildUrl(path6, params);
+      async *paginate(path7, params) {
+        let url = this.buildUrl(path7, params);
         while (url) {
           const res = await this.fetchImpl(url, { headers: this.headers() });
           if (!res.ok) {
@@ -186,9 +186,9 @@ var init_github = __esm({
           url = parseNextLink(res.headers.get("link"));
         }
       }
-      buildUrl(path6, params) {
+      buildUrl(path7, params) {
         const url = new URL(
-          `${this.baseUrl}/repos/${this.opts.owner}/${this.opts.repo}${path6}`
+          `${this.baseUrl}/repos/${this.opts.owner}/${this.opts.repo}${path7}`
         );
         url.searchParams.set("per_page", String(this.perPage));
         for (const [k, v] of Object.entries(params)) {
@@ -667,13 +667,13 @@ var require_textParsers = __commonJS({
         return fn(value);
       };
     }
-    function parseBool2(value) {
+    function parseBool(value) {
       if (value === null) return value;
       return value === "TRUE" || value === "t" || value === "true" || value === "y" || value === "yes" || value === "on" || value === "1";
     }
     function parseBoolArray(value) {
       if (!value) return null;
-      return array.parse(value, parseBool2);
+      return array.parse(value, parseBool);
     }
     function parseBaseTenInt(string) {
       return parseInt(string, 10);
@@ -808,7 +808,7 @@ var require_textParsers = __commonJS({
       register(26, parseInteger);
       register(700, parseFloat);
       register(701, parseFloat);
-      register(16, parseBool2);
+      register(16, parseBool);
       register(1082, parseDate);
       register(1114, parseDate);
       register(1184, parseDate);
@@ -1105,7 +1105,7 @@ var require_binaryParsers = __commonJS({
     var parseText = function(value) {
       return value.toString("utf8");
     };
-    var parseBool2 = function(value) {
+    var parseBool = function(value) {
       if (value === null) return null;
       return parseBits(value, 8) > 0;
     };
@@ -1117,7 +1117,7 @@ var require_binaryParsers = __commonJS({
       register(1700, parseNumeric);
       register(700, parseFloat32);
       register(701, parseFloat64);
-      register(16, parseBool2);
+      register(16, parseBool);
       register(1114, parseDate.bind(null, false));
       register(1184, parseDate.bind(null, true));
       register(1e3, parseArray);
@@ -3671,7 +3671,7 @@ var require_split2 = __commonJS({
 var require_helper = __commonJS({
   "../../node_modules/pgpass/lib/helper.js"(exports, module) {
     "use strict";
-    var path6 = __require("path");
+    var path7 = __require("path");
     var Stream2 = __require("stream").Stream;
     var split = require_split2();
     var util = __require("util");
@@ -3710,7 +3710,7 @@ var require_helper = __commonJS({
     };
     module.exports.getFileName = function(rawEnv) {
       var env = rawEnv || process.env;
-      var file = env.PGPASSFILE || (isWin ? path6.join(env.APPDATA || "./", "postgresql", "pgpass.conf") : path6.join(env.HOME || "./", ".pgpass"));
+      var file = env.PGPASSFILE || (isWin ? path7.join(env.APPDATA || "./", "postgresql", "pgpass.conf") : path7.join(env.HOME || "./", ".pgpass"));
       return file;
     };
     module.exports.usePgPass = function(stats, fname) {
@@ -3842,13 +3842,13 @@ var require_helper = __commonJS({
 var require_lib = __commonJS({
   "../../node_modules/pgpass/lib/index.js"(exports, module) {
     "use strict";
-    var path6 = __require("path");
+    var path7 = __require("path");
     var fs = __require("fs");
     var helper = require_helper();
     module.exports = function(connInfo, cb) {
       var file = helper.getFileName();
-      fs.stat(file, function(err, stat2) {
-        if (err || !helper.usePgPass(stat2, file)) {
+      fs.stat(file, function(err, stat3) {
+        if (err || !helper.usePgPass(stat3, file)) {
           return cb(void 0);
         }
         var st = fs.createReadStream(file);
@@ -5617,10 +5617,6 @@ function loadEnv() {
   }
   return { ...cachedFileVars, ...process.env };
 }
-function parseBool(v) {
-  if (v == null) return false;
-  return /^(1|true|yes|on)$/i.test(v.trim());
-}
 function loadConfig(env = loadEnv()) {
   return new Config(env);
 }
@@ -5667,7 +5663,6 @@ var init_config = __esm({
     };
     Config = class {
       environment;
-      cloudAiEnabled;
       databaseUrl;
       /** The tenant all connections are pinned to (RLS). DEFAULT_TENANT_ID in local mode. */
       tenantId;
@@ -5685,16 +5680,14 @@ var init_config = __esm({
           const tenant = env.BRAINJAR_TENANT_ID?.trim();
           if (!tenant) {
             throw new Error(
-              "ENVIRONMENT=saas requires BRAINJAR_TENANT_ID \u2014 every SaaS connection is tenant-scoped."
+              "ENVIRONMENT=saas requires BRAINJAR_TENANT_ID: every SaaS connection is tenant-scoped."
             );
           }
           this.databaseUrl = url;
           this.tenantId = tenant;
-          this.cloudAiEnabled = true;
         } else {
           this.databaseUrl = env.DATABASE_URL?.trim() || DEFAULT_DATABASE_URL;
           this.tenantId = DEFAULT_TENANT_ID;
-          this.cloudAiEnabled = parseBool(env.BRAINJAR_CLOUD_AI);
         }
       }
       /** Returns a secret, or throws an actionable error naming the env var and its purpose. */
@@ -5705,17 +5698,11 @@ var init_config = __esm({
           if (!value2) {
             throw new MissingSecretError(
               key,
-              `Missing ${CORPORATE_API_KEY_ENV}. In SaaS mode, ${spec.what} is billed to the central corporate account \u2014 set ${CORPORATE_API_KEY_ENV} in the backend environment.`
+              `Missing ${CORPORATE_API_KEY_ENV}. In SaaS mode, ${spec.what} is billed to the central corporate account: set ${CORPORATE_API_KEY_ENV} in the backend environment.`
             );
           }
           return value2;
         }
-        if (spec.cloudAi && !this.cloudAiEnabled) {
-          throw new MissingSecretError(
-            key,
-            `Cloud AI is disabled (BrainJar is local-first by default). To use ${spec.what}, set BRAINJAR_CLOUD_AI=true and ${spec.env}.`
-          );
-        }
         const value = this.#env[spec.env]?.trim();
         if (!value) {
           throw new MissingSecretError(
@@ -5738,7 +5725,6 @@ var init_config = __esm({
         return {
           environment: this.environment,
           tenantId: this.tenantId,
-          cloudAiEnabled: this.cloudAiEnabled,
           databaseUrl: redactUrlPassword(this.databaseUrl)
         };
       }
@@ -5753,12 +5739,14 @@ var init_config = __esm({
 });
 
 // ../../packages/shared/src/index.ts
+var PROSE_STYLE;
 var init_src = __esm({
   "../../packages/shared/src/index.ts"() {
     "use strict";
     init_memory();
     init_confidence();
     init_config();
+    PROSE_STYLE = "Never use em dashes (\u2014) or en dashes (\u2013) anywhere; use commas, colons, periods, or parentheses instead.";
   }
 });
 
@@ -5917,6 +5905,34 @@ var init_repository = __esm({
           [this.brainId]
         )).rows;
       }
+      /**
+       * Verified facts whose verified_scan evidence points at any of these repo-relative files —
+       * the re-verification trigger: a commit touched their proof, so their status may be stale.
+       */
+      async verifiedObjectsTouchingFiles(paths) {
+        if (paths.length === 0) return [];
+        return (await this.pool.query(
+          `SELECT DISTINCT ${MO_COLUMNS}, ${EV_AGG} FROM memory_object
+         WHERE brain_id = $1 AND status = 'active'
+           AND EXISTS (
+             SELECT 1 FROM evidence_link ev
+             WHERE ev.memory_object_id = memory_object.id
+               AND ev.artifact_kind = 'verified_scan'
+               AND EXISTS (
+                 SELECT 1 FROM unnest($2::text[]) AS p(path)
+                 WHERE ev.artifact_uri LIKE '%/' || p.path
+               )
+           )`,
+          [this.brainId, paths]
+        )).rows;
+      }
+      /** Replace a source's config (e.g. the GitHub repo/token a Brain reads from). */
+      async setSourceConfig(id, config) {
+        await this.pool.query(
+          "UPDATE source SET config = $3 WHERE id = $1 AND brain_id = $2",
+          [id, this.brainId, JSON.stringify(config)]
+        );
+      }
       async getSource(id) {
         return (await this.pool.query(
           "SELECT * FROM source WHERE id = $1 AND brain_id = $2",
@@ -6144,6 +6160,18 @@ var init_repository = __esm({
           [this.brainId, startIso, endIso, limit]
         )).rows;
       }
+      /**
+       * The newest substantive objects ("what are the latest ...?"). Entities are timeless
+       * structure, not news, so they are excluded along with summaries.
+       */
+      async newestObjects(limit = 8) {
+        return (await this.pool.query(
+          `SELECT ${MO_COLUMNS}, ${EV_AGG} FROM memory_object
+         WHERE brain_id = $1 AND type NOT IN ('summary', 'entity')
+         ORDER BY occurred_at DESC LIMIT $2`,
+          [this.brainId, limit]
+        )).rows;
+      }
       /** The daily_log (chronology rollup) for a given day, if one exists (one per active day). */
       async findDailyLog(date) {
         return (await this.pool.query(
@@ -6602,7 +6630,7 @@ var init_auth = __esm({
   "../../packages/db/src/auth.ts"() {
     "use strict";
     init_pool();
-    DEFAULT_TTL_HOURS = 24 * 7;
+    DEFAULT_TTL_HOURS = 24 * 90;
   }
 });
 
@@ -6653,10 +6681,24 @@ async function resolveSetting(key, envValue, pool2 = getPool()) {
     return void 0;
   }
 }
-var APP_SETTING_KEYS;
+async function resolveAnthropicApiKey(envKey, pool2 = getPool()) {
+  const fromEnv = envKey?.trim();
+  if (fromEnv) return fromEnv;
+  try {
+    const r = await pool2.query(
+      "SELECT value FROM app_setting WHERE key = $1",
+      [ANTHROPIC_KEY_SETTING]
+    );
+    return r.rows[0]?.value?.trim() || null;
+  } catch {
+    return null;
+  }
+}
+var APP_SETTING_KEYS, ANTHROPIC_KEY_SETTING;
 var init_settings = __esm({
   "../../packages/db/src/settings.ts"() {
     "use strict";
+    init_src();
     init_pool();
     APP_SETTING_KEYS = [
       "capture_mode",
@@ -6670,6 +6712,23 @@ var init_settings = __esm({
       "public_url"
       // this server's address as clients reach it (wizard command, MCP config)
     ];
+    ANTHROPIC_KEY_SETTING = "anthropic_api_key";
+  }
+});
+
+// ../../packages/db/src/notifications.ts
+var init_notifications = __esm({
+  "../../packages/db/src/notifications.ts"() {
+    "use strict";
+    init_pool();
+  }
+});
+
+// ../../packages/db/src/conversations.ts
+var init_conversations = __esm({
+  "../../packages/db/src/conversations.ts"() {
+    "use strict";
+    init_pool();
   }
 });
 
@@ -6685,6 +6744,8 @@ var init_src2 = __esm({
     init_team();
     init_rbac();
     init_settings();
+    init_notifications();
+    init_conversations();
   }
 });
 
@@ -6775,17 +6836,103 @@ var init_raw_store = __esm({
       async listForExtraction(brainId, limit = 500) {
         return (await this.pool.query(
           `SELECT * FROM raw_artifact
-         WHERE brain_id = $1 AND extracted_at IS NULL
+         WHERE brain_id = $1 AND extracted_at IS NULL AND kind <> 'source_file'
+         ORDER BY first_seen_at LIMIT $2`,
+          [brainId, limit]
+        )).rows;
+      }
+      /**
+       * Deep-scan queue: source files awaiting synthesis. Covers both fresh uploads (pending) and
+       * files stored while no AI key existed ('raw-source'), so adding a key later and re-running
+       * the scan picks them up.
+       */
+      /** How many shipped source files still await synthesis (pending or stored without a key). */
+      async sourceScanCount(brainId) {
+        return (await this.pool.query(
+          `SELECT count(*)::int AS n FROM raw_artifact
+           WHERE brain_id = $1 AND kind = 'source_file'
+             AND (extracted_at IS NULL OR extracted_by = 'raw-source')`,
+          [brainId]
+        )).rows[0]?.n ?? 0;
+      }
+      async listSourceFilesForScan(brainId, limit = 500) {
+        return (await this.pool.query(
+          `SELECT * FROM raw_artifact
+         WHERE brain_id = $1 AND kind = 'source_file'
+           AND (extracted_at IS NULL OR extracted_by = 'raw-source')
          ORDER BY first_seen_at LIMIT $2`,
           [brainId, limit]
         )).rows;
       }
-      async markExtracted(id) {
+      async markExtracted(id, by) {
         await this.pool.query(
-          "UPDATE raw_artifact SET extracted_at = now() WHERE id = $1",
-          [id]
+          "UPDATE raw_artifact SET extracted_at = now(), extracted_by = $2 WHERE id = $1",
+          [id, by]
         );
       }
+      /**
+       * The two numbers the Re-synthesize UI keys on. `pending` is the NORMAL pipeline (fresh
+       * artifacts not yet processed — never a reason to offer re-synthesis); `heuristic` is the
+       * backlog captured without AI, the only thing the button acts on.
+       */
+      async extractionState(brainId) {
+        const r = await this.pool.query(
+          `SELECT
+         count(*) FILTER (WHERE extracted_at IS NULL) AS pending,
+         count(*) FILTER (WHERE extracted_by = 'heuristic') AS heuristic
+       FROM raw_artifact WHERE brain_id = $1`,
+          [brainId]
+        );
+        return {
+          pending: Number(r.rows[0]?.pending ?? 0),
+          heuristic: Number(r.rows[0]?.heuristic ?? 0)
+        };
+      }
+      /**
+       * Prepare heuristically-extracted artifacts for AI re-extraction, atomically:
+       * their derived non-entity objects are deleted (evidence/object links cascade), stale
+       * entity evidence pointing at these artifacts is dropped (re-extraction re-adds it —
+       * entities themselves survive, resolveEntity dedupes them on the re-run), and the
+       * artifacts return to the pipeline (extracted_at/by cleared). Returns how many artifacts
+       * were re-queued — 0 means nothing to do, which is also what makes a double click a no-op.
+       */
+      async clearForResynthesis(brainId) {
+        const client = await this.pool.connect();
+        try {
+          await client.query("BEGIN");
+          const ids = (await client.query(
+            `SELECT id FROM raw_artifact
+           WHERE brain_id = $1 AND extracted_by = 'heuristic' FOR UPDATE`,
+            [brainId]
+          )).rows.map((r) => r.id);
+          if (ids.length === 0) {
+            await client.query("ROLLBACK");
+            return 0;
+          }
+          await client.query(
+            `DELETE FROM memory_object WHERE type <> 'entity' AND id IN (
+           SELECT memory_object_id FROM evidence_link
+           WHERE raw_artifact_id = ANY($1::uuid[]))`,
+            [ids]
+          );
+          await client.query(
+            "DELETE FROM evidence_link WHERE raw_artifact_id = ANY($1::uuid[])",
+            [ids]
+          );
+          await client.query(
+            `UPDATE raw_artifact SET extracted_at = NULL, extracted_by = NULL
+         WHERE id = ANY($1::uuid[])`,
+            [ids]
+          );
+          await client.query("COMMIT");
+          return ids.length;
+        } catch (e) {
+          await client.query("ROLLBACK");
+          throw e;
+        } finally {
+          client.release();
+        }
+      }
     };
   }
 });
@@ -8261,12 +8408,12 @@ var init_path = __esm({
   "../../node_modules/@anthropic-ai/sdk/internal/utils/path.mjs"() {
     init_error();
     EMPTY = /* @__PURE__ */ Object.freeze(/* @__PURE__ */ Object.create(null));
-    createPathTagFunction = (pathEncoder = encodeURIPath) => function path6(statics, ...params) {
+    createPathTagFunction = (pathEncoder = encodeURIPath) => function path7(statics, ...params) {
       if (statics.length === 1)
         return statics[0];
       let postPath = false;
       const invalidSegments = [];
-      const path7 = statics.reduce((previousValue, currentValue, index) => {
+      const path8 = statics.reduce((previousValue, currentValue, index) => {
         if (/[?#]/.test(currentValue)) {
           postPath = true;
         }
@@ -8283,7 +8430,7 @@ var init_path = __esm({
         }
         return previousValue + currentValue + (index === params.length ? "" : encoded);
       }, "");
-      const pathOnly = path7.split(/[?#]/, 1)[0];
+      const pathOnly = path8.split(/[?#]/, 1)[0];
       const invalidSegmentPattern = /(?<=^|\/)(?:\.|%2e){1,2}(?=\/|$)/gi;
       let match;
       while ((match = invalidSegmentPattern.exec(pathOnly)) !== null) {
@@ -8304,10 +8451,10 @@ var init_path = __esm({
         }, "");
         throw new AnthropicError(`Path parameters result in path with invalid segments:
 ${invalidSegments.map((e) => e.error).join("\n")}
-${path7}
+${path8}
 ${underline}`);
       }
-      return path7;
+      return path8;
     };
     path2 = /* @__PURE__ */ createPathTagFunction(encodeURIPath);
   }
@@ -11293,9 +11440,9 @@ var init_client = __esm({
       makeStatusError(status, error, message, headers) {
         return APIError.generate(status, error, message, headers);
       }
-      buildURL(path6, query, defaultBaseURL) {
+      buildURL(path7, query, defaultBaseURL) {
         const baseURL = !__classPrivateFieldGet(this, _BaseAnthropic_instances, "m", _BaseAnthropic_baseURLOverridden).call(this) && defaultBaseURL || this.baseURL;
-        const url = isAbsoluteURL(path6) ? new URL(path6) : new URL(baseURL + (baseURL.endsWith("/") && path6.startsWith("/") ? path6.slice(1) : path6));
+        const url = isAbsoluteURL(path7) ? new URL(path7) : new URL(baseURL + (baseURL.endsWith("/") && path7.startsWith("/") ? path7.slice(1) : path7));
         const defaultQuery = this.defaultQuery();
         if (!isEmptyObj(defaultQuery)) {
           query = { ...defaultQuery, ...query };
@@ -11326,24 +11473,24 @@ var init_client = __esm({
        */
       async prepareRequest(request, { url, options }) {
       }
-      get(path6, opts) {
-        return this.methodRequest("get", path6, opts);
+      get(path7, opts) {
+        return this.methodRequest("get", path7, opts);
       }
-      post(path6, opts) {
-        return this.methodRequest("post", path6, opts);
+      post(path7, opts) {
+        return this.methodRequest("post", path7, opts);
       }
-      patch(path6, opts) {
-        return this.methodRequest("patch", path6, opts);
+      patch(path7, opts) {
+        return this.methodRequest("patch", path7, opts);
       }
-      put(path6, opts) {
-        return this.methodRequest("put", path6, opts);
+      put(path7, opts) {
+        return this.methodRequest("put", path7, opts);
       }
-      delete(path6, opts) {
-        return this.methodRequest("delete", path6, opts);
+      delete(path7, opts) {
+        return this.methodRequest("delete", path7, opts);
       }
-      methodRequest(method, path6, opts) {
+      methodRequest(method, path7, opts) {
         return this.request(Promise.resolve(opts).then((opts2) => {
-          return { method, path: path6, ...opts2 };
+          return { method, path: path7, ...opts2 };
         }));
       }
       request(options, remainingRetries = null) {
@@ -11447,8 +11594,8 @@ var init_client = __esm({
         }));
         return { response, options, controller, requestLogID, retryOfRequestLogID, startTime };
       }
-      getAPIList(path6, Page2, opts) {
-        return this.requestAPIList(Page2, { method: "get", path: path6, ...opts });
+      getAPIList(path7, Page2, opts) {
+        return this.requestAPIList(Page2, { method: "get", path: path7, ...opts });
       }
       requestAPIList(Page2, options) {
         const request = this.makeRequest(options, null, void 0);
@@ -11535,8 +11682,8 @@ var init_client = __esm({
       }
       async buildRequest(inputOptions, { retryCount = 0 } = {}) {
         const options = { ...inputOptions };
-        const { method, path: path6, query, defaultBaseURL } = options;
-        const url = this.buildURL(path6, query, defaultBaseURL);
+        const { method, path: path7, query, defaultBaseURL } = options;
+        const url = this.buildURL(path7, query, defaultBaseURL);
         if ("timeout" in options)
           validatePositiveInteger("timeout", options.timeout);
         options.timeout = options.timeout ?? this.timeout;
@@ -11652,11 +11799,11 @@ var init_sdk = __esm({
 
 // ../../packages/ingest/src/llm.ts
 function rationaleModelId(env = process.env) {
-  return env.BRAINJAR_RATIONALE_MODEL?.trim() || "claude-opus-4-8";
+  return env.BRAINJAR_RATIONALE_MODEL?.trim() || "claude-opus-5";
 }
-function anthropicClientFromConfig(config, model) {
+function anthropicClient(apiKey, model) {
   return new AnthropicClient({
-    apiKey: config.requireSecret("anthropicApiKey"),
+    apiKey,
     model: model?.trim() || rationaleModelId()
   });
 }
@@ -11685,6 +11832,24 @@ var init_llm = __esm({
         });
         return res.content.map((b) => b.type === "text" ? b.text : "").join("").trim();
       }
+      /** Same completion, but yielded as text deltas for surfaces that render tokens live (Ask chat). */
+      async *completeStream({
+        system,
+        prompt,
+        maxTokens = 1024
+      }) {
+        const stream = this.client.messages.stream({
+          model: this.model,
+          max_tokens: maxTokens,
+          ...system ? { system } : {},
+          messages: [{ role: "user", content: prompt }]
+        });
+        for await (const event of stream) {
+          if (event.type === "content_block_delta" && event.delta.type === "text_delta") {
+            yield event.delta.text;
+          }
+        }
+      }
     };
   }
 });
@@ -11694,12 +11859,14 @@ var SYSTEM;
 var init_rationale_extractor = __esm({
   "../../packages/ingest/src/rationale-extractor.ts"() {
     "use strict";
+    init_src();
     init_heuristic_extractor();
     SYSTEM = [
       "You extract the WHY behind a software artifact for a memory system.",
-      "Given an artifact (commit, PR, review comment, or agent session), explain in 1-3 sentences why it exists",
-      "\u2014 the decision, intent, or business reason it reflects. Infer from the evidence; do not invent specifics.",
-      "Respond with only the rationale sentences. No preamble, no headings, no quotes."
+      "Given an artifact (commit, PR, review comment, or agent session), explain in 1-3 sentences why it exists:",
+      "the decision, intent, or business reason it reflects. Infer from the evidence; do not invent specifics.",
+      "Respond with only the rationale sentences. No preamble, no headings, no quotes.",
+      PROSE_STYLE
     ].join(" ");
   }
 });
@@ -11719,6 +11886,14 @@ var init_impact = __esm({
   }
 });
 
+// ../../packages/ingest/src/commit-links.ts
+var init_commit_links = __esm({
+  "../../packages/ingest/src/commit-links.ts"() {
+    "use strict";
+    init_src2();
+  }
+});
+
 // ../../packages/ingest/src/extraction.ts
 var init_extraction = __esm({
   "../../packages/ingest/src/extraction.ts"() {
@@ -11728,6 +11903,9 @@ var init_extraction = __esm({
     init_raw_store();
     init_signal();
     init_impact();
+    init_commit_links();
+    init_entities();
+    init_rationale_extractor();
   }
 });
 
@@ -11797,6 +11975,9 @@ var init_structure = __esm({
     SKIP_DIRS = /* @__PURE__ */ new Set([
       "node_modules",
       ".git",
+      // Parallel agent worktrees live inside the repo; scanning them creates duplicate entities
+      // with .claude/worktrees/... paths that pollute retrieval.
+      ".claude",
       ".next",
       "dist",
       "build",
@@ -11814,6 +11995,7 @@ var init_product_facts = __esm({
     "use strict";
     init_src2();
     init_entities();
+    init_src();
     PRODUCT_FACTS_SYSTEM = [
       "You read a raw coding-agent session and extract durable PRODUCT facts for a documentation system.",
       "Output ONLY a JSON array (no prose, no code fences). Each item:",
@@ -11822,8 +12004,9 @@ var init_product_facts = __esm({
       "A 'feature' is user- or developer-facing capability (a button, a page, an endpoint, a command).",
       "A 'decision' is an architectural/design choice with a reason.",
       "Capture what was BUILT or CHANGED, where it lives (files/symbols), how it behaves, why, and any",
-      "configuration it exposes. Prefer a few high-signal facts over many trivial ones. Invent nothing \u2014",
-      "if the session only debugged or explored, return []."
+      "configuration it exposes. Prefer a few high-signal facts over many trivial ones. Invent nothing:",
+      "if the session only debugged or explored, return [].",
+      PROSE_STYLE
     ].join(" ");
   }
 });
@@ -11840,24 +12023,18 @@ var CONSOLIDATE_SYSTEM;
 var init_compress = __esm({
   "../../packages/ingest/src/compress.ts"() {
     "use strict";
+    init_src();
     CONSOLIDATE_SYSTEM = [
       "You compress a raw multi-turn coding-agent session for a memory system.",
       "Produce a single high-density summary with exactly three sections: Intent, Failed Attempts, Resolution.",
       "Preserve 100% of the architectural reasoning and the chronological order of events.",
       "Keep every distinct error message (once) and every rejected approach with why it failed.",
-      "Drop all pleasantries, tool chatter, and restated code. Target at most half the input length."
+      "Drop all pleasantries, tool chatter, and restated code. Target at most half the input length.",
+      PROSE_STYLE
     ].join(" ");
   }
 });
 
-// ../../packages/ingest/src/commit-links.ts
-var init_commit_links = __esm({
-  "../../packages/ingest/src/commit-links.ts"() {
-    "use strict";
-    init_src2();
-  }
-});
-
 // ../../packages/ingest/src/transcript.ts
 var init_transcript = __esm({
   "../../packages/ingest/src/transcript.ts"() {
@@ -12643,6 +12820,7 @@ var init_doku = __esm({
   "../../packages/ingest/src/doku.ts"() {
     "use strict";
     init_src2();
+    init_src();
     init_doku_model();
     DOKU_FILENAME = "BRAINJAR_DOKU.md";
     DOKU_DIR = "brainjar-doku";
@@ -12655,14 +12833,83 @@ var init_doku = __esm({
     );
     isModule = (o) => o.type === "entity" && o.body.kind === "module";
     moduleName = (o) => String(o.body.name ?? o.title);
-    isRepoPath = (name) => !/^([a-zA-Z]:[/\\]|[/\\])/.test(name) && !name.includes("node_modules");
+    isRepoPath = (name) => !/^([a-zA-Z]:[/\\]|[/\\])/.test(name) && !name.includes("node_modules") && !name.includes(".claude/");
     normPath = (p) => p.replace(/\\/g, "/").replace(/^\.\//, "").toLowerCase();
     NARRATIVE_SYSTEM = [
       "You write the introduction for one feature's documentation page, for a mixed audience",
       "(developers, product managers, leadership). You are given the page's structured facts as",
       "markdown. Respond with 1-3 short plain-text paragraphs that explain the feature naturally:",
       "what it is, why it exists, how it evolved. Keep every fact; invent nothing. No headings,",
-      "no lists, no markdown syntax \u2014 prose only."
+      "no lists, no markdown syntax, prose only.",
+      PROSE_STYLE
+    ].join(" ");
+  }
+});
+
+// ../../packages/ingest/src/deepscan.ts
+var SOURCE_FACTS_SYSTEM;
+var init_deepscan = __esm({
+  "../../packages/ingest/src/deepscan.ts"() {
+    "use strict";
+    init_src2();
+    init_src();
+    init_raw_store();
+    init_product_facts();
+    init_doku();
+    SOURCE_FACTS_SYSTEM = [
+      "You read SOURCE FILES of a software project (each introduced by a '=== path ===' header) and extract durable PRODUCT facts for a documentation system.",
+      "Output ONLY a JSON array (no prose, no code fences). Each item:",
+      '{ "kind": "feature"|"decision", "title": string, "summary": string,',
+      '  "where": string[], "behavior"?: string, "rationale"?: string, "config"?: string[] }.',
+      "A 'feature' is user- or developer-facing capability (a page, an endpoint, a command, a job).",
+      "A 'decision' is an architectural/design choice whose reason is visible in the code or comments.",
+      "Describe what the code DOES today, where it lives (use the given paths in 'where'), how it",
+      "behaves, why it is built that way when stated, and any configuration it exposes.",
+      "Prefer a few high-signal facts per batch over many trivial ones. Invent nothing:",
+      "if the files carry no product knowledge (pure plumbing), return [].",
+      PROSE_STYLE
+    ].join(" ");
+  }
+});
+
+// ../../packages/ingest/src/github-app.ts
+var init_github_app = __esm({
+  "../../packages/ingest/src/github-app.ts"() {
+    "use strict";
+  }
+});
+
+// ../../packages/ingest/src/repo-reader.ts
+var init_repo_reader = __esm({
+  "../../packages/ingest/src/repo-reader.ts"() {
+    "use strict";
+    init_structure();
+    init_github_app();
+  }
+});
+
+// ../../packages/ingest/src/verify-agent.ts
+var VERIFY_SYSTEM;
+var init_verify_agent = __esm({
+  "../../packages/ingest/src/verify-agent.ts"() {
+    "use strict";
+    init_sdk();
+    init_src2();
+    init_src();
+    init_product_facts();
+    init_doku();
+    init_llm();
+    VERIFY_SYSTEM = [
+      "You are BrainJar's verification agent. You read a software repository and record durable product knowledge, with one iron rule: prose is never truth, code is.",
+      "Documentation (READMEs, docs) gives you HYPOTHESES. Before recording any fact, read the implementing code with read_file and check it.",
+      "Record via record_fact with the honest verification status:",
+      "- confirmed: you read the implementation and it does this today.",
+      "- contradicted: the docs claim one thing, the code does another. Record what the CODE does as the fact; put the outdated claim in `contradicts`.",
+      "- documented-only: claimed in prose, but you could not find it in the code. Record it as documented intent, phrase the summary accordingly.",
+      "Always fill `where` with the repo paths you actually read for this fact.",
+      "Work highest-signal first: what the product is, its main capabilities, its architecture decisions. A few verified facts beat many shallow ones.",
+      "When you are done (or told the budget is exhausted), reply with a short plain-text summary instead of more tool calls.",
+      PROSE_STYLE
     ].join(" ");
   }
 });
@@ -12711,6 +12958,10 @@ var init_src3 = __esm({
     init_summaries();
     init_doku();
     init_doku_model();
+    init_deepscan();
+    init_repo_reader();
+    init_github_app();
+    init_verify_agent();
     init_scheduler();
     registerConnector(
       "github",
@@ -12883,6 +13134,7 @@ async function init(argv) {
     `
 \u2705  Brain "${name}" is live \u2014 ${final.objects} memory objects.
     Open the web app and visit Docs. Approve the hook in Claude Code when prompted.
+    Optional: \`npx brainjar deepscan\` reads the repo's key files so the docs start full.
 `
   );
   return 0;
@@ -13226,6 +13478,174 @@ var init_doctor = __esm({
   }
 });
 
+// ../../packages/cli/src/deepscan.ts
+var deepscan_exports = {};
+__export(deepscan_exports, {
+  deepscanMain: () => deepscanMain,
+  selectDeepScanFiles: () => selectDeepScanFiles
+});
+import { readdir as readdir3, readFile as readFile6, stat as stat2 } from "node:fs/promises";
+import path6 from "node:path";
+function selectDeepScanFiles(candidates, graph, maxFiles = MAX_FILES) {
+  const picked = [];
+  const has = /* @__PURE__ */ new Set();
+  const take = (p) => {
+    if (!has.has(p) && picked.length < maxFiles) {
+      has.add(p);
+      picked.push(p);
+    }
+  };
+  for (const bucket of BUCKET_CAPS) {
+    let n = 0;
+    const ordered = [...candidates].sort().reverse();
+    for (const p of ordered) {
+      if (n >= bucket.cap) break;
+      if (bucket.test(p) && !has.has(p)) {
+        take(p);
+        n++;
+      }
+    }
+  }
+  const inDegree = /* @__PURE__ */ new Map();
+  for (const e of graph.edges)
+    inDegree.set(e.to, (inDegree.get(e.to) ?? 0) + 1);
+  const central = [...inDegree.entries()].sort((a, b) => b[1] - a[1]).map(([p]) => p);
+  for (const p of central) {
+    if (picked.length >= maxFiles) break;
+    take(p);
+  }
+  return picked;
+}
+async function walkCandidates(root) {
+  const entries = await readdir3(root, { recursive: true, withFileTypes: true });
+  const out = [];
+  for (const e of entries) {
+    if (!e.isFile()) continue;
+    const parent = e.parentPath ?? e.path ?? root;
+    const rel = path6.relative(root, path6.join(parent, e.name)).split(path6.sep).join("/");
+    if (rel.split("/").some((seg) => SKIP_DIRS.has(seg))) continue;
+    out.push(rel);
+  }
+  return out;
+}
+async function deepscanMain() {
+  const root = await repoRoot();
+  const project = await captureApi(`/capture/project?root=${encodeURIComponent(root)}`);
+  if (!project.brainId || !project.sourceId) {
+    console.error(
+      `brainjar: no Brain is connected to ${root} \u2014 run \`npx brainjar init\` first`
+    );
+    process.exitCode = 1;
+    return;
+  }
+  console.error("brainjar: selecting key files\u2026");
+  const [candidates, graph] = await Promise.all([
+    walkCandidates(root),
+    scanStructure(root)
+  ]);
+  const files = selectDeepScanFiles(candidates, graph);
+  const artifacts = [];
+  for (const rel of files) {
+    try {
+      const abs = path6.join(root, rel);
+      const info = await stat2(abs);
+      if (info.size > RAW_SKIP_BYTES) continue;
+      const content = await readFile6(abs, "utf8");
+      artifacts.push({
+        source: "deepscan",
+        kind: "source_file",
+        externalId: `file:${rel}`,
+        uri: rel,
+        sha: contentHash(content),
+        title: rel,
+        body: content.slice(0, FILE_CAP),
+        author: null,
+        createdAt: info.mtime.toISOString(),
+        updatedAt: info.mtime.toISOString(),
+        payload: { path: rel, bytes: info.size }
+      });
+    } catch {
+    }
+  }
+  let shipped = 0;
+  let queued = 0;
+  let chunk = [];
+  let chunkChars = 0;
+  const flush = async (final) => {
+    if (chunk.length === 0 && !final) return;
+    const r = await captureApi(
+      "/capture/artifacts",
+      {
+        method: "POST",
+        body: {
+          root,
+          sourceId: project.sourceId,
+          artifacts: chunk,
+          // Only the final chunk kicks the server-side synthesis: one scan per run.
+          deepscanFinal: final
+        }
+      }
+    );
+    shipped += chunk.length;
+    queued += r.deepscanQueued ?? 0;
+    chunk = [];
+    chunkChars = 0;
+  };
+  for (const a of artifacts) {
+    if (chunk.length >= CHUNK_FILES || chunkChars + a.body.length > CHUNK_CHARS)
+      await flush(false);
+    chunk.push(a);
+    chunkChars += a.body.length;
+  }
+  await flush(true);
+  console.error(
+    `brainjar: deep scan shipped ${shipped} files (${queued} queued for synthesis).`
+  );
+  console.error(
+    "brainjar: synthesis runs on the server in the background \u2014 the bell in the app reports the outcome. Unchanged files were deduplicated automatically."
+  );
+}
+var FILE_CAP, CHUNK_CHARS, CHUNK_FILES, MAX_FILES, RAW_SKIP_BYTES, BUCKET_CAPS;
+var init_deepscan2 = __esm({
+  "../../packages/cli/src/deepscan.ts"() {
+    "use strict";
+    init_src3();
+    init_api_client();
+    FILE_CAP = 16e3;
+    CHUNK_CHARS = 65e4;
+    CHUNK_FILES = 40;
+    MAX_FILES = 60;
+    RAW_SKIP_BYTES = 2e5;
+    BUCKET_CAPS = [
+      // READMEs first — they outrank everything else the repo says about itself.
+      { cap: 4, test: (p) => /(^|\/)readme[^/]*$/i.test(p) },
+      // Markdown docs.
+      { cap: 6, test: (p) => /^docs\//i.test(p) && p.endsWith(".md") },
+      // Manifests: what the project IS and depends on.
+      {
+        cap: 10,
+        test: (p) => /(^|\/)(package\.json|pyproject\.toml|go\.mod|cargo\.toml|composer\.json|gemfile)$/i.test(
+          p
+        ) && p.split("/").length <= 3
+      },
+      // Schema + deployment: migrations, compose files, example env.
+      {
+        cap: 12,
+        test: (p) => /(^|\/)migrations\//.test(p) && p.endsWith(".sql") || /(^|\/)(docker-compose[^/]*\.ya?ml|\.env\.example|schema\.[^/]+)$/i.test(
+          p
+        )
+      },
+      // Entry points and API surfaces by name.
+      {
+        cap: 12,
+        test: (p) => /(^|\/)(index|main|app|server|cli|routes?|api)\.(ts|tsx|js|py|go|rb)$/.test(
+          p
+        )
+      }
+    ];
+  }
+});
+
 // ../../packages/cli/src/unhook.ts
 var unhook_exports = {};
 __export(unhook_exports, {
@@ -13275,11 +13695,11 @@ async function dokuMain(argv = []) {
       return;
     }
     let llm;
-    const config = loadConfig();
-    if (config.cloudAiEnabled) {
+    const key = await resolveAnthropicApiKey(loadEnv().ANTHROPIC_API_KEY);
+    if (key) {
       try {
-        llm = anthropicClientFromConfig(
-          config,
+        llm = anthropicClient(
+          key,
           await resolveSetting(
             "rationale_model",
             loadEnv().BRAINJAR_RATIONALE_MODEL
@@ -13329,6 +13749,7 @@ Usage:
   brainjar constraints [--hook]      Known failed approaches (SessionStart hook entrypoint)
   brainjar failure --hook            Near-live failure capture (PostToolUseFailure entrypoint)
   brainjar doctor [--no-sync]        Check the capture pipeline end to end
+  brainjar deepscan                  Read the repo's key files into the Brain (AI synthesis server-side)
   brainjar unhook                    Remove all BrainJar hooks (data stays)
   brainjar doku [--export]           Rebuild the documentation pages (--export dumps them to files)
   brainjar help                      Show this help
@@ -13361,6 +13782,11 @@ async function main() {
       await doctorMain2(argv);
       break;
     }
+    case "deepscan": {
+      const { deepscanMain: deepscanMain2 } = await Promise.resolve().then(() => (init_deepscan2(), deepscan_exports));
+      await deepscanMain2();
+      break;
+    }
     case "unhook": {
       const { unhookMain: unhookMain2 } = await Promise.resolve().then(() => (init_unhook(), unhook_exports));
       await unhookMain2();

package/package.json

@@ -1,9 +1,9 @@
 {
   "name": "brainjar",
-  "version": "0.1.0",
+  "version": "0.2.0",
   "description": "BrainJar CLI: connect a repository to a BrainJar server (capture, hooks, MCP)",
-  "license": "SEE LICENSE ON https://ponytail.dev",
-  "author": "ponytailhq <ops@ponytail.dev>",
+  "license": "SEE LICENSE ON https://brainjar.dev",
+  "author": "Dietrich Gebert",
   "homepage": "https://brainjar.dev",
   "type": "module",
   "bin": {