Skip to content

lemon07r/Vera

Repository files navigation

vera

Vera

License: MIT Rust GitHub release Languages

Install Guide · Features · Query Guide · Benchmarks · How It Works · Models · Supported Languages

Vector Enhanced Reranking Agent

Code search that combines BM25 keyword matching, vector similarity, and cross-encoder reranking. Supports 64 languages (60 with tree-sitter parsing), runs locally, returns structured results with file paths, line ranges, symbol metadata, and relevance scores.

Quick Start

1. Install

bunx @vera-ai/cli install   # or: npx -y @vera-ai/cli install / uvx vera-ai install

2. Set up models (pick one)

vera setup                       # Interactive wizard (auto-detects your hardware)
vera setup --api                 # API mode: works everywhere, no GPU needed (recommended)
vera setup --onnx-jina-coreml    # Apple Silicon (M1/M2/M3/M4)
vera setup --onnx-jina-cuda      # NVIDIA GPU
vera setup --onnx-jina-rocm      # AMD GPU (ROCm, Linux)
vera setup --onnx-jina-openvino  # Intel GPU (OpenVINO, Linux)
vera setup --onnx-jina-directml  # DirectX 12 GPU (Windows)

3. Index and search

vera index .
vera search "authentication logic"

What Sets Vera Apart

Cross-encoder reranking Most tools stop at retrieval. Vera scores query-candidate pairs jointly, lifting MRR@10 from 0.28 to 0.60.
Single binary, 64 languages One static binary with 60 tree-sitter grammars compiled in. No Python, no language servers, no per-language toolchains.
Built-in code intelligence Call graph analysis, reference finding, dead code detection, and project overview, all from the same index.
Token-efficient for agents Returns symbol-bounded chunks, not entire files. 75-95% fewer tokens on typical queries.

Vera started after weeks of working on Pampax, a project I forked because it and other similar tools were missing what I wanted. I kept running into deep-rooted bugs, less-than-ideal design decisions, and thought I could build something better from the ground up. Every design choice comes from careful research, learning from other projects, benchmarking and evaluation. Take a look at the full feature list to see everything Vera can do.

Installation

bunx @vera-ai/cli install   # or: npx -y @vera-ai/cli install / uvx vera-ai install

Pick Your Backend

Vera itself is always local: the index lives in .vera/ per project, config and models in $XDG_DATA_HOME/vera (or ~/.vera for existing installs). The backend choice only affects where embeddings and reranking run.

You have Run this What happens
Not sure vera setup Interactive wizard auto-detects your hardware
Any hardware vera setup --api Models run remotely via any OpenAI-compatible API. No GPU needed. Recommended.
Apple Silicon (M1/M2/M3/M4) vera setup --onnx-jina-coreml Downloads local models, uses CoreML GPU acceleration
NVIDIA GPU vera setup --onnx-jina-cuda Downloads local models, uses CUDA. Fastest local option
AMD GPU (Linux) vera setup --onnx-jina-rocm Downloads local models, uses ROCm
Intel GPU (Linux) vera setup --onnx-jina-openvino Downloads local models, uses OpenVINO
DirectX 12 GPU (Windows) vera setup --onnx-jina-directml Downloads local models, uses DirectML

API mode works with any OpenAI-compatible endpoint and needs no local compute. Local mode downloads two curated ONNX models and auto-detects your GPU; a GPU is recommended since CPU-only indexing is slow. After the first index, vera update . only re-embeds changed files, so incremental updates are fast on any backend. Full details: docs/models.md.

For step-by-step instructions, API provider options, Docker, building from source, and troubleshooting, see the full Installation Guide.

MCP server
vera mcp   # or: bunx @vera-ai/cli mcp / uvx vera-ai mcp

Exposes search_code, get_stats, get_overview, and regex_search tools. search_code auto-indexes and starts a file watcher on first use if no index exists.

Usage

vera search "authentication logic"
vera search "error handling" --lang rust
vera search "routes" --path "src/**/*.ts"
vera search "handler" --type function --limit 5
vera search "config loading" --deep              # query decomposition + parallel search (or iterative symbol-following)
vera search "auth" --compact                     # signatures only, broad exploration

Update the index after code changes: vera update .

Keep the index fresh automatically: vera watch .

See the query guide for tips on writing effective queries.

More commands
# Code intelligence
vera grep "fn\s+main"              # regex search over indexed files
vera references foo                # find all callers of symbol 'foo'
vera references foo --callees      # find what 'foo' calls
vera dead-code                     # find functions with no callers
vera overview                      # project summary: languages, entry points, hotspots

# Index management
vera index .                       # index current project
vera update .                      # re-index changed files
vera watch .                       # auto-update on file changes (Ctrl-C to stop)
vera stats                         # index statistics

# Setup and diagnostics
vera doctor                        # diagnose setup issues
vera doctor --probe                # deeper read-only ONNX probe
vera repair                        # re-fetch missing assets
vera upgrade                       # inspect binary update plan
vera config                        # show current configuration
vera backend                       # manage ONNX runtime and model backend

# Agent skills
vera agent install                 # interactive: choose scope + agents
vera agent install --client all    # non-interactive: all agents, global
vera agent status                  # check skill installation status
vera agent sync                    # refresh stale skill installs
vera agent remove                  # pick installs to remove

# Cleanup
vera uninstall                     # removes config dir, skill files, PATH shim

Output Format

Defaults to markdown codeblocks (the most token-efficient format for AI agents):

```src/auth/login.rs:42-68 function:authenticate
pub fn authenticate(credentials: &Credentials) -> Result<Token> { ... }
```

Use --json for compact JSON, --raw for verbose human-readable output, or --timing for per-stage pipeline durations.

Excluding Files

Vera respects .gitignore by default. Create a .veraignore file (gitignore syntax) for more control, or use --exclude flags. Details: docs/features.md.

Benchmarks

21-task benchmark across ripgrep, flask, fastify, and turborepo:

Metric ripgrep cocoindex ColGREP (149M) Vera
Recall@5 0.28 0.37 0.67 0.78
MRR@10 0.26 0.35 0.62 0.91
nDCG@10 0.29 0.52 0.56 0.84

Full methodology and version history: docs/benchmarks.md.

Configure Your AI Agent

vera agent install installs the Vera skill for your coding agents and offers to add a usage snippet to your project's AGENTS.md, CLAUDE.md, COPILOT.md, or editor rules file. Installed agents start preselected in the interactive picker, deselecting one removes its existing Vera skill install, and stale installs can be refreshed in one step before you enter the full picker.

Alternatively, install the Vera skill with the skills CLI:

npx skills add lemon07r/Vera

If you skipped the prompt or want to add it manually:

## Code Search

Use Vera before opening many files or running broad text search when you need to find where logic lives or how a feature works.

- `vera search "query"` for semantic code search. Describe behavior: "JWT validation", not "auth".
- `vera grep "pattern"` for exact text or regex
- `vera references <symbol>` for callers and callees
- `vera overview` for a project summary (languages, entry points, hotspots)
- `vera search --deep "query"` for query decomposition + parallel search with weighted fusion
- Narrow results with `--lang`, `--path`, `--type`, or `--scope docs`
- `vera watch .` to auto-update the index, or `vera update .` after edits (`vera index .` if `.vera/` is missing)
- For detailed usage, query patterns, and troubleshooting, read the Vera skill file installed by `vera agent install`

Contributing

See CONTRIBUTING.md.

About

Local code search combining BM25, vector similarity, and cross-encoder reranking. Parses 60+ languages with tree-sitter, runs entirely offline, and returns structured results with file paths, line ranges, and symbol metadata. Built in Rust.

Topics

Resources

License

Contributing

Stars

Watchers

Forks

Packages

 
 
 

Contributors