CodexA

Appearance

Upgrade Guide ​

Upgrading to v0.5.0 ​

v0.5.0 is a major release spanning 12 phases (31-42) that brings a native Rust search engine, RAG pipeline, advanced search features, multi-editor support, cross-language intelligence, and multi-agent orchestration.

Install ​

bash
pip install --upgrade codexa

# Or with ML extras
pip install --upgrade "codexa[ml]"

What's New ​

Rust Search Engine Core (Phase 32) ​

  • Native codexa-core Rust crate via PyO3 for transparent acceleration
  • HNSW approximate nearest-neighbour search with memory-mapped persistence
  • AST-aware chunker splitting code at function/class boundaries (10 languages)
  • BM25 keyword index, parallel file scanner, optional ONNX embedding inference
  • Graceful fallback: all Rust components are optional

RAG Pipeline (Phase 31) ​

  • 4-stage Retrieval-Augmented Generation: Retrieve, Deduplicate, Re-rank, Assemble
  • Cross-encoder re-ranking, token-aware budgets, source citations
  • Configurable strategies: semantic, keyword, hybrid, multi

Search & Grep Output (Phase 33) ​

New flags on codexa search and codexa grep:

bash
# JSONL streaming (ideal for AI agents and pipelines)
codexa search "auth flow" --jsonl
codexa grep "TODO" --jsonl

# Show relevance scores
codexa search "error handling" --scores

# Control snippet length or omit snippets entirely
codexa search "database" --snippet-length 100
codexa search "database" --no-snippet

# Exclude files by glob, include gitignored files
codexa search "config" --exclude "*.test.*" --no-ignore
codexa grep "FIXME" --exclude "vendor/*" -L

Incremental Indexing (Phase 34) ​

bash
# Index a single file without full scan
codexa index --add src/new_module.py

# Inspect a file's index metadata (content hash, chunks, vectors)
codexa index --inspect src/auth.py
  • Model-consistency guard: Detects if the embedding model changed since last index and warns before corrupting vectors
  • Ctrl+C safety: Partial index is saved on interrupt; next run resumes

Tantivy Full-Text Engine (Phase 35) ​

Optional Rust-native Tantivy full-text search engine. Requires building codexa-core with the tantivy-backend feature:

bash
cd codexa-core
maturin develop --release --features tantivy-backend

Check availability:

python
from semantic_code_intelligence.rust_backend import use_tantivy
print(use_tantivy())  # True if compiled with tantivy-backend

MCP Server v2 (Phase 36) ​

  • Cursor-based pagination on semantic_search, keyword_search, hybrid_search MCP tools
  • codexa --serve shorthand to start the MCP server in one flag
  • Claude Desktop auto-config:
bash
codexa mcp --claude-config

Search Shorthands & Distribution (Phase 37) ​

bash
# Mode shorthands
codexa search "query" --hybrid    # equivalent to --mode hybrid
codexa search "query" --sem       # equivalent to --mode semantic

# .codexaignore is auto-created on first index with sensible defaults

PyInstaller spec included at codexa.spec for single-binary distribution.

Model Hub & Hot-Swap (Phase 38) ​

bash
# Switch embedding model with automatic re-index
codexa index --switch-model jina-code

# Download with integrity verification
codexa models download bge-small --verify

# Benchmark models with memory metrics
codexa models benchmark
  • Per-model vector subdirectories for multi-model index support

Pre-built Wheels & Distribution (Phase 39) ​

  • CI-built wheels for Linux (x86_64, aarch64), macOS (universal2), Windows (x64)
  • Scoop and Chocolatey package manifests
  • Standalone PyInstaller binaries via GitHub Releases
  • Updated Docker image with Rust extensions

Code Editor Plugins (Phase 40) ​

First-class integration for 9 editors, all sharing the same MCP/bridge protocol:

  • Zed -- extension with context servers and language servers
  • JetBrains -- IntelliJ/PyCharm/WebStorm plugin with bridge HTTP client
  • Neovim -- Lua plugin with telescope.nvim picker and floating preview
  • Vim -- Vimscript plugin with quickfix integration
  • Sublime Text -- command palette, quick panel, output panel
  • Emacs -- helm/ivy completion, grep-mode results
  • Helix -- languages.toml configuration guide
  • Eclipse -- plugin descriptor and setup guide
  • Cursor/Windsurf -- documented MCP setup configs

Multi-Agent Orchestration (Phase 41) ​

  • Thread-safe SessionManager for concurrent AI agent sessions with TTL cleanup
  • Shared discovery pool for coordinated context across agents
  • Semantic diff: AST-level rename, move, signature, body, and cosmetic change detection
  • RAG-grounded code generator
  • Bridge session endpoints: /sessions, /sessions/create, /sessions/close

Cross-Language Intelligence (Phase 42) ​

  • FFI pattern detection (Python-Rust, Python-C, JS-WASM, Java-JNI)
  • Polyglot dependency graphs across language boundaries
  • Language-aware search boosting with configurable boost factor
  • Universal multi-language call graph

Breaking Changes ​

None. All new flags and features are additive. Existing CLI invocations work unchanged.

Migration Checklist ​

  1. pip install --upgrade codexa
  2. Re-index to pick up the latest features: codexa index --force
  3. (Optional) Build with Tantivy: maturin develop --release --features tantivy-backend
  4. (Optional) Set up Claude Desktop: codexa mcp --claude-config
  5. (Optional) Try model switching: codexa index --switch-model jina-code

Previous Versions ​

v0.4.5 ​

  • RAG Pipeline (Phase 31): 4-stage Retrieve, Deduplicate, Re-rank, Assemble
  • Cross-encoder re-ranking, token-aware budgets, source citations

v0.4.4 ​

  • Model Profiles: fast, balanced, precise with auto-detection
  • codexa init --profile, codexa models profiles, codexa models benchmark

v0.4.3 ​

  • Bundled tree-sitter grammars, exclude_files config
  • Install tiers: pip install codexa (lightweight) vs pip install "codexa[ml]"

v0.4.0 ​

  • First stable public release
  • 39 CLI commands, 13 AI agent tools, 22 plugin hooks