v9.5.0 — Config Validation & Language Support

What's Changed — v9.5.0: Config Validation & Language Support

Features

• config validate CLI command with line-number tracking and fix suggestions
• config migrate and config diff for schema version upgrades
• Wizard warns on invalid config before server start
• Object Pascal AST-aware ingestion (.pas/.pp/.dpr/.dpk) with function/class extraction
• OpenCode installer fixes: singular dirs, agent frontmatter conversion, path rewriting, permission pre-auth, idempotent install
• Nested storage.postgres.* config key validation with POSTGRES_KNOWN_FIELDS allowlist
• pool_timeout added to PostgresConfig (default 30s), wired through create_async_engine
• MODE_SUPPORT_MATRIX for explicit unsupported mode reporting in benchmarks
• BENCHMARKS.md with real baseline latency data (5-10ms p50, 95-188 QPS on Chroma)

Bug Fixes

• 120s start timeout (was 30s), state_dir path resolution, ChromaDB telemetry suppression, Gemini google-genai migration

Tests

• 11 OCDI requirement tests, config validation/migration tests, 6 nested postgres validation tests, 36 benchmark helper tests

PyPI Packages

• agent-brain-rag: https://pypi.org/project/agent-brain-rag/9.5.0/
• agent-brain-cli: https://pypi.org/project/agent-brain-cli/9.5.0/

Installation

pip install agent-brain-rag==9.5.0 agent-brain-cli==9.5.0

# With GraphRAG support
pip install "agent-brain-rag[graphrag]==9.5.0"

Full Changelog: v9.4.1...v9.5.0

v9.4.1

Full Changelog: v9.3.1...v9.4.1

v9.4.0

v9.4.0 Released

This release is now live on PyPI:

• agent-brain-rag 9.4.0: https://pypi.org/project/agent-brain-rag/9.4.0/
• agent-brain-cli 9.4.0: https://pypi.org/project/agent-brain-cli/9.4.0/

Full Changelog: v9.3.0...v9.4.0

v9.3.0 - LangExtract Multi-Provider Graph Extraction + 12-Step Config Wizard

Phase 35: LangExtract Document Graph Extractor with multi-provider support (Gemini, OpenAI, Claude, Ollama). Phase 34: 12-step config wizard. GRAPH_USE_LLM_EXTRACTION defaults to False. 14 new unit tests.

v9.2.0

What's Changed

Features

• feat: add Codex named adapter with AGENTS.md generation (#119)
• feat: add generic skill-runtime converter and parser extensions (#119)
• feat: add docs, integration tests, and quality gate for v9.1 (#119)
• feat(ci): add YAML frontmatter lint to before-push and PR QA gate (#120)

Bug Fixes

• fix(plugin): quote YAML frontmatter descriptions containing colons (#120)
• fix(ci): install pyyaml before running YAML frontmatter lint step (#120)

Documentation

• docs: v9.2.0 Documentation Accuracy Audit — audited all 71 docs with last_validated metadata (#120)
• docs: comprehensive updates to USER_GUIDE, API_REFERENCE, CONFIGURATION, PLUGIN_GUIDE, PROVIDER_CONFIGURATION (#120)
• docs: define milestone v9.1.0 requirements and roadmap (#119)
• docs: archive v8.0 and v9.0 milestones (#119)

Other Changes

• chore: close out v9.1.0 milestone in planning docs (#119)

About Agent Brain

Agent Brain provides intelligent document indexing and semantic search for AI agents:

• Semantic Search: Natural language queries via OpenAI embeddings
• Keyword Search (BM25): Traditional keyword matching with TF-IDF
• GraphRAG: Knowledge graph retrieval for relationship-aware queries
• Hybrid Search: Best of vector + keyword approaches
• Pluggable Providers: Choose your embedding and summarization providers
• Multi-Runtime: Install plugin for Claude Code, OpenCode, Gemini CLI, or Codex

PyPI Packages

• agent-brain-rag: https://pypi.org/project/agent-brain-rag/9.2.0/
• agent-brain-cli: https://pypi.org/project/agent-brain-cli/9.2.0/

Installation

pip install agent-brain-rag==9.2.0 agent-brain-cli==9.2.0

# With GraphRAG support
pip install "agent-brain-rag[graphrag]==9.2.0"

Documentation

• User Guide
• Developer Guide
• All Releases

Full Changelog: v9.0.0...v9.2.0

v9.0.0

What's Changed

Features

• Multi-runtime support for Claude, OpenCode, and Gemini (#118)
  • agent-brain install-agent now supports --runtime claude|opencode|gemini
  • Converts Agent Brain skill/config files to each runtime's native format
  • Adds OpenCode (agents.md) and Gemini CLI (GEMINI.md) converter support

Full Changelog

v8.0.0...v9.0.0

v8.0.0

What's Changed

• v8.0: Phases 15, 16, 19 — File Watcher, Embedding Cache, Plugin Updates by @RichardHightower in #116
• feat: phases 24-25 — setup permissions bootstrap + wizard coverage gaps by @RichardHightower in #117

Full Changelog: v7.0.0...v8.0.0

v7.0.0 - Manifest Tracking, Chunk Eviction & Content Injection

What's New in v7.0.0

This major release delivers three significant capabilities: manifest tracking, chunk eviction, and content injection. Together these features complete the v7.0 milestone for Agent Brain.

Highlights

Phase 14: Manifest Tracking & Chunk Eviction

• ManifestTracker service — tracks which chunks belong to which source files, enabling precise lifecycle management
• ChunkEvictionService — automatically removes stale chunks when source files are re-indexed or deleted
• Pipeline integration — manifest tracking and eviction wired into the indexing pipeline end-to-end
• CLI eviction summary — displays eviction results after indexing operations
• Full test coverage — unit tests for manifest and eviction services, integration tests for pipeline wiring

Phase 13: Content Injection Pipeline

• ContentInjector service — programmatic injection of content into the index without file-based ingestion
• API router extension — new /index/inject endpoint for direct content injection
• CLI inject command — agent-brain inject for injecting content from the command line
• Extended models — IndexRequest and JobRecord models extended to support injection workflows
• Injector protocol documentation (INJECT-08)

Phase 12: Folder Management & File Type Presets

• Folder management — organize and manage indexed folders
• File type presets — pre-configured file type filters for common use cases
• include_types pipeline — filter indexed files by type during ingestion

Full Changelog

Features

• feat(14-02): CLI eviction summary display and integration tests (b32fb92)
• feat(14-02): wire ManifestTracker and ChunkEvictionService into indexing pipeline (6f482cb)
• feat(14-01): extend storage_paths, JobRecord, add unit tests for manifest and eviction (1caf26b)
• feat(14-01): add ManifestTracker and ChunkEvictionService foundation services (447fc1c)
• feat(13-02): implement inject CLI command and extend DocServeClient (2916384)
• feat(13-01): Integrate ContentInjector into pipeline, job worker, and API router (4bd9ae3)
• feat(13-01): ContentInjector service and extend IndexRequest/JobRecord models (9842acf)
• feat(12): folder management, file type presets, and include_types pipeline (b3a5be4)

Bug Fixes

• fix(13): revise plans based on checker feedback (77bf131)
• fix(test): accept **kwargs in fake_stat mocks for Python 3.11+ compat (8c35327)

Style

• style(14): apply Black formatting and add planning docs (3f3e8e7)
• style(14-02): fix ruff import order and lint issues in test files (bf045df)
• style(13): apply Black formatting to content injection files (11c8369)

Testing

• test(14): complete UAT - 7 passed, 0 issues (38b0ea9)

Documentation

• docs(phase-14): complete phase execution (b0be3e5)
• docs(14-02): complete manifest tracking pipeline wiring plan (df4c3de)
• docs(14-01): complete ManifestTracker and ChunkEvictionService foundation plan (3cc9631)
• docs(14): create phase plan for manifest tracking and chunk eviction (8251001)
• docs(phase-13): complete phase execution (35d6f10)
• docs(13-02): complete CLI inject command plan (e01ce62)
• docs(13-02): create injector protocol documentation (1ced5e3)
• docs(13-01): complete content injection pipeline plan summary (00c2365)
• docs(13): create phase 13 content injection pipeline plans (cd4737e)
• docs(13): research phase 13 content injection pipeline (aaea853)

Other

• chore: complete v7.0 milestone — archive and evolve PROJECT.md (4727d90)

---

PyPI Packages

Packages will be available on PyPI in ~5 minutes:

• agent-brain-rag
• agent-brain-cli

Install / Upgrade

uv pip install agent-brain-rag==7.0.0
uv pip install agent-brain-cli==7.0.0

v6.0.3 — Fix PostgreSQL Integration Bugs

What's Changed

Fixes 6 bugs discovered during PostgreSQL backend integration testing.

Bug Fixes

• [CRITICAL] Job verification uses empty ChromaDB instead of active backend — job_worker.py now calls storage_backend.get_count() instead of vector_store.get_count(), which was reading from an empty ChromaDB when PostgreSQL was active, causing all indexing jobs to falsely report FAILED
• [MEDIUM] asyncpg and sqlalchemy missing from dependencies — Moved from optional to required dependencies so PostgreSQL backend works out of the box
• [LOW] .docx files crash when docx2txt missing — .docx files are now gracefully skipped with a warning when docx2txt is not installed
• [LOW] Status endpoint shows 0 documents — Root cause was the job verification bug (Bug 1); status endpoint was already correct
• [LOW] .claude/ directory not excluded from indexing — Added .claude/ and .claude-plugin/ to DEFAULT_EXCLUDE_PATTERNS
• [COSMETIC] Config not hot-reloaded — Added restart note to config command docs

Testing

• 772 tests passing (686 server + 86 CLI)
• 74% server coverage, 54% CLI coverage
• 5 new tests for job worker PostgreSQL verification

Full Changelog: v6.0.2...v6.0.3

v6.0.1 — PostgreSQL SQL Cast Fix

v6.0.1 — PostgreSQL SQL Cast Fix

Patch release fixing a SQL syntax error that broke all PostgreSQL backend operations.

PyPI Packages

• agent-brain-rag 6.0.1
• agent-brain-cli 6.0.1

Bug Fix

SQLAlchemy's text() treats :name as bind parameters, so PostgreSQL ::type cast syntax (e.g., :embedding::vector) produces invalid SQL — the bind parameter gets replaced but the ::type suffix remains as a literal.

Affected operations: All PostgreSQL upsert and search operations (vector search, keyword search, hybrid search, metadata filtering).

Root cause: 8 instances of ::cast syntax in SQL strings with bind parameters across 3 files.

Changes

File	Fix
keyword_ops.py	:metadata::jsonb → CAST(:metadata AS jsonb)
vector_ops.py	:embedding::vector → CAST(:embedding AS vector)
vector_ops.py	:query_embedding::vector → CAST(:query_embedding AS vector) (x2)
vector_ops.py	:filter::jsonb → CAST(:filter AS jsonb)
backend.py	:filter::jsonb → CAST(:filter AS jsonb)
connection.py	Clamp pool.overflow() with max(0, ...) for accurate pool metrics
test_storage_config.py	Isolate test from CI DATABASE_URL environment variable

Impact

• ChromaDB users (default): Not affected
• PostgreSQL users: All vector/keyword/hybrid operations were broken in 6.0.0

Upgrade

pip install --upgrade agent-brain-rag==6.0.1 agent-brain-cli==6.0.1

Full Changelog: v6.0.0...v6.0.1