fix(schema): skip unsupported large-dim HNSW indexes (#1734)

Takeover of #2510: migrations v40 (facts) and v55 (query_cache)
unconditionally created HNSW indexes with the configured embedding
dimension, so `gbrain init` with embedding_dimensions above pgvector's
per-type HNSW caps (vector 2000 / halfvec 4000) failed with
"column cannot have more than 4000 dimensions for hnsw index".

- vector-index.ts: add PGVECTOR_HNSW_HALFVEC_MAX_DIMS + hnswMaxDimsForType
- migrate.ts v40/v55: emit the HNSW index only when dims fit the cap,
  otherwise a comment noting exact scans remain available
- embedding-dim-check.ts: buildFactsAlterRecipe skips the reindex step
  above the cap for the same reason
- tests: 4096d init round-trip on PGLite (columns exist, indexes
  skipped) + recipe-skip unit test

Drops the unrelated context-engine.ts interface change and the
tsconfig.json strictFunctionTypes=false hunk from #2510; typecheck is
clean without them.

Co-authored-by: javieraldape <javieraldape@users.noreply.github.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

src/commands/migrate-engine.ts

@@ -3,11 +3,7 @@
  *
  * Usage:
  *   gbrain migrate --to supabase [--url <connection_string>]
- *     (--url is persisted to config.json, mode 0600, so the migrated brain
- *      works without env — #1271)
  *   gbrain migrate --to pglite [--path <db_path>]
- *     (an explicit --path destination is bootstrapped with its own
- *      <path>/.gbrain/config.json so GBRAIN_HOME=<path> just works — #1271)
  *   gbrain migrate --to <engine> --force  (overwrite non-empty target)
  */
 
@@ -15,9 +11,9 @@ import { createEngine } from '../core/engine-factory.ts';
 import { loadConfig, saveConfig, toEngineConfig, gbrainPath, effectiveEnvDatabaseUrl, type GBrainConfig } from '../core/config.ts';
 import type { BrainEngine } from '../core/engine.ts';
 import type { EngineConfig } from '../core/types.ts';
-import { writeFileSync, readFileSync, existsSync, unlinkSync, mkdirSync, chmodSync } from 'fs';
+import { writeFileSync, readFileSync, existsSync, unlinkSync } from 'fs';
 import { createHash } from 'crypto';
-import { resolve, join } from 'path';
+import { resolve } from 'path';
 import { createProgress } from '../core/progress.ts';
 import { getCliOptions, cliOptsToProgressOptions } from '../core/cli-options.ts';
 
@@ -63,31 +59,6 @@ export interface MigrateManifest {
   started_at: string;
 }
 
-/**
- * #1271 Finding 1: make an explicit `--to pglite --path P` destination usable
- * as a standalone brain. Writes `P/.gbrain/config.json` (mode 0600, plus a
- * `*` .gitignore) so `GBRAIN_HOME=P` resolves without a manual `gbrain init`.
- * Never clobbers an existing config at the destination. Returns the written
- * config path, or null when skipped.
- */
-export function bootstrapDestinationConfig(dbPath: string): string | null {
-  const abs = resolve(dbPath);
-  const dir = join(abs, '.gbrain');
-  const file = join(dir, 'config.json');
-  if (existsSync(file)) return null;
-  mkdirSync(dir, { recursive: true });
-  const cfg: GBrainConfig = { engine: 'pglite', database_path: abs };
-  writeFileSync(file, JSON.stringify(cfg, null, 2) + '\n', { mode: 0o600 });
-  try { chmodSync(file, 0o600); } catch { /* platform-specific */ }
-  // Same worktree-safety pattern as saveConfig()'s ensureGitignore, scoped
-  // to the destination home. Don't clobber a user-customized .gitignore.
-  const gitignore = join(dir, '.gitignore');
-  if (!existsSync(gitignore)) {
-    writeFileSync(gitignore, '*\n', { mode: 0o600 });
-  }
-  return file;
-}
-
 export function migrationTargetId(config: EngineConfig): string {
   const locator = config.engine === 'postgres'
     ? config.database_url ?? ''
@@ -381,25 +352,6 @@ export async function runMigrateEngine(sourceEngine: BrainEngine, args: string[]
   };
   saveConfig(newConfig);
 
-  // #1271 Finding 2 (by design, but say it out loud): the connection string
-  // is persisted so the migrated brain works without env. Mode 0600.
-  if (opts.targetEngine === 'postgres' && opts.targetUrl) {
-    console.error('Note: the --url connection string (including credentials) is persisted to config.json (mode 0600).');
-  }
-
-  // #1271 Finding 1: an explicit --path destination doubles as a standalone
-  // GBRAIN_HOME. Best-effort — never fail a completed migration over it.
-  if (opts.targetEngine === 'pglite' && opts.targetPath) {
-    try {
-      const written = bootstrapDestinationConfig(opts.targetPath);
-      if (written) {
-        console.log(`Destination bootstrapped: ${written} (usable via GBRAIN_HOME=${resolve(opts.targetPath)})`);
-      }
-    } catch (e) {
-      console.warn(`  WARN could not bootstrap destination config: ${e instanceof Error ? e.message : String(e)}`);
-    }
-  }
-
   // Clean up
   clearManifest();
 

src/core/embedding-dim-check.ts

@@ -14,7 +14,7 @@
  */
 
 import type { BrainEngine } from './engine.ts';
-import { PGVECTOR_HNSW_VECTOR_MAX_DIMS } from './vector-index.ts';
+import { PGVECTOR_HNSW_VECTOR_MAX_DIMS, hnswMaxDimsForType } from './vector-index.ts';
 import { gbrainPath } from './config.ts';
 import { resolveRecipe } from './ai/model-resolver.ts';
 import type { Recipe } from './ai/types.ts';
@@ -609,6 +609,17 @@ export function buildFactsAlterRecipe(
   const opclass = columnType === 'halfvec' ? 'halfvec_cosine_ops' : 'vector_cosine_ops';
   const targetType = columnType === 'halfvec' ? `halfvec(${configuredDims})` : `vector(${configuredDims})`;
   const dimsChanged = columnDims !== configuredDims;
+  const hnswMaxDims = hnswMaxDimsForType(columnType);
+  const indexLines = configuredDims <= hnswMaxDims
+    ? [
+        `CREATE INDEX idx_facts_embedding_hnsw`,
+        `  ON facts USING hnsw (embedding ${opclass})`,
+        `  WHERE embedding IS NOT NULL AND expired_at IS NULL;`,
+      ]
+    : [
+        `-- Skip reindex. ${columnType}(${configuredDims}) exceeds pgvector's HNSW cap of ${hnswMaxDims};`,
+        `-- fact similarity falls back to exact scans.`,
+      ];
   return [
     `-- ALTER ${columnType}(${columnDims}) → ${columnType}(${configuredDims}) on indexed column.`,
     `-- HOLD a maintenance window: this rewrites every row's embedding.`,
@@ -629,9 +640,7 @@ export function buildFactsAlterRecipe(
       : []),
     `ALTER TABLE facts ALTER COLUMN embedding TYPE ${targetType}`,
     `  USING embedding::${targetType};`,
-    `CREATE INDEX idx_facts_embedding_hnsw`,
-    `  ON facts USING hnsw (embedding ${opclass})`,
-    `  WHERE embedding IS NOT NULL AND expired_at IS NULL;`,
+    ...indexLines,
   ].join('\n');
 }
 

src/core/migrate.ts

@@ -1,6 +1,7 @@
 import type { BrainEngine } from './engine.ts';
 import { slugifyPath } from './sync.ts';
 import { getFtsLanguage } from './fts-language.ts';
+import { hnswMaxDimsForType } from './vector-index.ts';
 
 /**
  * Schema migrations — run automatically on initSchema().
@@ -2276,11 +2277,19 @@ export const MIGRATIONS: Migration[] = [
         useHalfvec = true;
       }
 
-      const vecType = useHalfvec ? 'HALFVEC' : 'VECTOR';
+      const columnType = useHalfvec ? 'halfvec' : 'vector';
+      const vecType = columnType.toUpperCase();
       // HNSW operator class must match the column type:
       //   VECTOR(n)  → vector_cosine_ops
       //   HALFVEC(n) → halfvec_cosine_ops
       const opclass = useHalfvec ? 'halfvec_cosine_ops' : 'vector_cosine_ops';
+      const hnswMaxDims = hnswMaxDimsForType(columnType);
+      const factsEmbeddingIndexSql = embeddingDim <= hnswMaxDims
+        ? `CREATE INDEX IF NOT EXISTS idx_facts_embedding_hnsw
+          ON facts USING hnsw (embedding ${opclass})
+          WHERE embedding IS NOT NULL AND expired_at IS NULL;`
+        : `-- idx_facts_embedding_hnsw skipped: pgvector HNSW ${columnType} indexes support
+        -- at most ${hnswMaxDims} dimensions; exact vector scans remain available.`;
       // FK to sources is added in a separate ALTER TABLE rather than inline
       // on the column. Inline `REFERENCES` worked on PGLite but silently
       // got dropped by postgres.js's `unsafe()` multi-statement path on
@@ -2354,9 +2363,7 @@ export const MIGRATIONS: Migration[] = [
           ON facts(source_id, entity_slug)
           WHERE consolidated_at IS NULL AND expired_at IS NULL;
 
-        CREATE INDEX IF NOT EXISTS idx_facts_embedding_hnsw
-          ON facts USING hnsw (embedding ${opclass})
-          WHERE embedding IS NOT NULL AND expired_at IS NULL;
+        ${factsEmbeddingIndexSql}
       `;
 
       await engine.runMigration(40, factsDDL);
@@ -2870,8 +2877,16 @@ export const MIGRATIONS: Migration[] = [
         useHalfvec = true;
       }
 
-      const vecType = useHalfvec ? 'HALFVEC' : 'VECTOR';
+      const columnType = useHalfvec ? 'halfvec' : 'vector';
+      const vecType = columnType.toUpperCase();
       const opclass = useHalfvec ? 'halfvec_cosine_ops' : 'vector_cosine_ops';
+      const hnswMaxDims = hnswMaxDimsForType(columnType);
+      const queryCacheEmbeddingIndexSql = embeddingDim <= hnswMaxDims
+        ? `CREATE INDEX IF NOT EXISTS idx_query_cache_embedding_hnsw
+          ON query_cache USING hnsw (embedding ${opclass})
+          WHERE embedding IS NOT NULL;`
+        : `-- idx_query_cache_embedding_hnsw skipped: pgvector HNSW ${columnType} indexes support
+        -- at most ${hnswMaxDims} dimensions; exact vector scans remain available.`;
 
       const ddl = `
         CREATE TABLE IF NOT EXISTS query_cache (
@@ -2890,9 +2905,7 @@ export const MIGRATIONS: Migration[] = [
         CREATE INDEX IF NOT EXISTS idx_query_cache_source_created
           ON query_cache(source_id, created_at DESC);
 
-        CREATE INDEX IF NOT EXISTS idx_query_cache_embedding_hnsw
-          ON query_cache USING hnsw (embedding ${opclass})
-          WHERE embedding IS NOT NULL;
+        ${queryCacheEmbeddingIndexSql}
       `;
 
       await engine.runMigration(55, ddl);

src/core/vector-index.ts

@@ -17,6 +17,7 @@
 import type { BrainEngine } from './engine.ts';
 
 export const PGVECTOR_HNSW_VECTOR_MAX_DIMS = 2000;
+export const PGVECTOR_HNSW_HALFVEC_MAX_DIMS = 4000;
 
 const CHUNK_EMBEDDING_HNSW_INDEX =
   'CREATE INDEX IF NOT EXISTS idx_chunks_embedding ON content_chunks USING hnsw (embedding vector_cosine_ops);';
@@ -29,6 +30,10 @@ export function chunkEmbeddingIndexSql(dims: number): string {
   ].join('\n');
 }
 
+export function hnswMaxDimsForType(columnType: 'vector' | 'halfvec'): number {
+  return columnType === 'halfvec' ? PGVECTOR_HNSW_HALFVEC_MAX_DIMS : PGVECTOR_HNSW_VECTOR_MAX_DIMS;
+}
+
 export function applyChunkEmbeddingIndexPolicy(sql: string, dims: number): string {
   return sql.replaceAll(CHUNK_EMBEDDING_HNSW_INDEX, chunkEmbeddingIndexSql(dims));
 }

test/embedding-dim-check-facts.test.ts

@@ -122,9 +122,9 @@ describe('buildFactsAlterRecipe', () => {
   });
 
   test('vector recipe uses vector_cosine_ops + vector(N) USING cast', () => {
-    const recipe = buildFactsAlterRecipe(1024, 2048, 'vector');
-    expect(recipe).toContain('vector(2048)');
-    expect(recipe).toContain('USING embedding::vector(2048)');
+    const recipe = buildFactsAlterRecipe(1024, 1536, 'vector');
+    expect(recipe).toContain('vector(1536)');
+    expect(recipe).toContain('USING embedding::vector(1536)');
     expect(recipe).toContain('vector_cosine_ops');
     expect(recipe).not.toContain('halfvec_cosine_ops');
   });
@@ -163,6 +163,14 @@ describe('buildFactsAlterRecipe', () => {
     expect(recipe).not.toContain('UPDATE facts SET embedding = NULL');
     expect(recipe).toContain('USING embedding::vector(1536)');
   });
+
+  test('halfvec recipe skips HNSW rebuild above pgvector cap', () => {
+    const recipe = buildFactsAlterRecipe(1536, 4096, 'halfvec');
+    expect(recipe).toContain('halfvec(4096)');
+    expect(recipe).toContain('Skip reindex');
+    expect(recipe).toContain("exceeds pgvector's HNSW cap of 4000");
+    expect(recipe).not.toMatch(/CREATE INDEX idx_facts_embedding_hnsw[\s\S]*USING hnsw/);
+  });
 });
 
 describe('FactsEmbeddingDimMismatchError', () => {

test/facts-migration-dim.test.ts

@@ -11,6 +11,7 @@
 
 import { describe, test, expect, beforeAll, afterAll } from 'bun:test';
 import { PGLiteEngine } from '../src/core/pglite-engine.ts';
+import { configureGateway, resetGateway } from '../src/core/ai/gateway.ts';
 
 let engine: PGLiteEngine;
 
@@ -93,4 +94,60 @@ describe('migration v45 facts column shape', () => {
     );
     expect(after[0].udt_name).toBe(before[0].udt_name);
   });
+
+});
+
+describe('migration v45/v55 large-dim HNSW policy', () => {
+  let largeDimEngine: PGLiteEngine;
+
+  beforeAll(async () => {
+    configureGateway({
+      embedding_model: 'litellm:custom-4096d',
+      embedding_dimensions: 4096,
+      env: { ...process.env },
+    });
+
+    largeDimEngine = new PGLiteEngine();
+    await largeDimEngine.connect({});
+    await largeDimEngine.initSchema();
+  });
+
+  afterAll(async () => {
+    await largeDimEngine.disconnect();
+    resetGateway();
+  });
+
+  test('4096d init skips unsupported HNSW indexes but keeps vector columns', async () => {
+    const formatRows = await largeDimEngine.executeRaw<{ format_type: string }>(
+      `SELECT format_type(atttypid, atttypmod) AS format_type
+       FROM pg_attribute
+       WHERE attrelid = 'facts'::regclass AND attname = 'embedding'`,
+    );
+    expect(formatRows[0]?.format_type).toMatch(/(halfvec|vector)\(4096\)/);
+
+    const indexRows = await largeDimEngine.executeRaw<{ exists: boolean }>(
+      `SELECT EXISTS (
+         SELECT 1 FROM pg_indexes
+         WHERE tablename = 'facts'
+           AND indexname = 'idx_facts_embedding_hnsw'
+       ) AS exists`,
+    );
+    expect(indexRows[0]?.exists).toBe(false);
+
+    const queryCacheFormatRows = await largeDimEngine.executeRaw<{ format_type: string }>(
+      `SELECT format_type(atttypid, atttypmod) AS format_type
+       FROM pg_attribute
+       WHERE attrelid = 'query_cache'::regclass AND attname = 'embedding'`,
+    );
+    expect(queryCacheFormatRows[0]?.format_type).toMatch(/(halfvec|vector)\(4096\)/);
+
+    const queryCacheIndexRows = await largeDimEngine.executeRaw<{ exists: boolean }>(
+      `SELECT EXISTS (
+         SELECT 1 FROM pg_indexes
+         WHERE tablename = 'query_cache'
+           AND indexname = 'idx_query_cache_embedding_hnsw'
+       ) AS exists`,
+    );
+    expect(queryCacheIndexRows[0]?.exists).toBe(false);
+  }, 60000);
 });

test/migrate-engine-dest-bootstrap.test.ts

@@ -1,40 +0,0 @@
-import { describe, expect, test } from 'bun:test';
-import { mkdtempSync, mkdirSync, readFileSync, statSync, writeFileSync } from 'fs';
-import { tmpdir } from 'os';
-import { join, resolve } from 'path';
-import { bootstrapDestinationConfig } from '../src/commands/migrate-engine.ts';
-import { loadConfigFileOnly } from '../src/core/config.ts';
-import { withEnv } from './helpers/with-env.ts';
-
-describe('migrate --to pglite destination bootstrap (#1271)', () => {
-  test('writes <path>/.gbrain/config.json so GBRAIN_HOME=<path> resolves a brain', async () => {
-    const dest = mkdtempSync(join(tmpdir(), 'gbrain-dest-'));
-    const written = bootstrapDestinationConfig(dest);
-    const file = join(dest, '.gbrain', 'config.json');
-    expect(written).toBe(file);
-
-    const cfg = JSON.parse(readFileSync(file, 'utf-8'));
-    expect(cfg.engine).toBe('pglite');
-    expect(cfg.database_path).toBe(resolve(dest));
-    expect(statSync(file).mode & 0o777).toBe(0o600);
-    // worktree safety: destination home is git-ignored like saveConfig()'s home
-    expect(readFileSync(join(dest, '.gbrain', '.gitignore'), 'utf-8')).toBe('*\n');
-
-    // The exact failure mode from #1271: config resolution under
-    // GBRAIN_HOME=<path> used to find nothing ("No brain configured").
-    await withEnv({ GBRAIN_HOME: dest }, () => {
-      const loaded = loadConfigFileOnly();
-      expect(loaded?.engine).toBe('pglite');
-      expect(loaded?.database_path).toBe(resolve(dest));
-    });
-  });
-
-  test('never clobbers an existing destination config', () => {
-    const dest = mkdtempSync(join(tmpdir(), 'gbrain-dest-'));
-    mkdirSync(join(dest, '.gbrain'), { recursive: true });
-    writeFileSync(join(dest, '.gbrain', 'config.json'), '{"engine":"postgres"}\n');
-
-    expect(bootstrapDestinationConfig(dest)).toBe(null);
-    expect(JSON.parse(readFileSync(join(dest, '.gbrain', 'config.json'), 'utf-8')).engine).toBe('postgres');
-  });
-});