Vera

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 optional cross-encoder reranking. Supports 65 languages (61 with tree-sitter parsing), runs locally, returns structured results with file paths, line ranges, symbol metadata, and relevance scores.

See What's New for release highlights from v1.0 onward, including search-quality measurements, local model changes, agent workflows, performance work, and reliability fixes.

1. Install

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

2. Set up and index (pick one)

vera setup                                  # Interactive, indexes this project by default
vera setup --potion-code --index .          # Default local model
vera setup --api --index .                  # Remote API mode, prompts for endpoint + key
vera setup --onnx-jina-coreml --index .     # Apple Silicon (M1/M2/M3/M4)
vera setup --onnx-jina-cuda --index .       # NVIDIA GPU
vera setup --onnx-jina-rocm --index .       # AMD GPU (ROCm, Linux)
vera setup --onnx-jina-openvino --index .   # Intel GPU (OpenVINO, Linux)
vera setup --onnx-jina-directml --index .   # DirectX 12 GPU (Windows)

The interactive vera setup wizard offers presets for OpenAI, Jina, Voyage, and Qwen via OpenRouter; the Qwen preset uses qwen/qwen3-embedding-8b + qwen/qwen3-reranker-8b via https://openrouter.ai/api/v1 with a single shared key and generic reranker protocol.

3. Search

vera search "authentication logic"

If the current project has no index, interactive search offers to create one. JSON and non-interactive searches still return the missing-index error.

The default local embedding model is minishlab/potion-code-16M-v2. It runs locally on CPU on any supported machine; no GPU or ONNX Runtime needed. Jina ONNX and CodeRankEmbed are opt-in alternatives.

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.

Use the quick start above if you just want to get going. This section helps you pick the right backend.

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

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.

Pick the vera setup flag that matches your hardware from the quick start above. The full hardware-to-command matrix, step-by-step instructions, API provider options, Docker, and building from source live in the Installation Guide.

API mode works with any OpenAI-compatible endpoint and needs no local compute. Use vera setup --api --yes with EMBEDDING_MODEL_* variables for non-interactive setup. The Qwen preset (qwen/qwen3-embedding-8b + qwen/qwen3-reranker-8b via https://openrouter.ai/api/v1) needs only one shared key and configures the generic reranker protocol automatically. Jina ONNX and CodeRankEmbed are opt-in alternatives. Reranking is opt-in and disabled by default. After the first index, vera update . only re-embeds changed files, so incremental updates are fast on any backend. Full details: docs/models.md.

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

vera search "authentication logic"
vera update .

vera search "error handling" --lang rust
vera search "routes" --path "src/**/*.ts" --path "tests/**/*.ts"
vera search "handler" --type function --limit 5
vera search "OAuth token refresh" "JWT expiry handling" "auth middleware"
vera search "config" --intent "find where database connection strings are loaded"
vera search "config loading" --deep
vera search "auth" --compact
vera search "token validation" --changed
vera search "config loading" --base origin/main
vera structural definitions parse_config
vera structural env DATABASE_URL
vera structural routes --path "src/**/*.ts"
vera structural impls Loader
vera references parse_config --changed

Repeat --path to match any of several file path patterns. Path patterns use OR semantics; other filters still combine with AND semantics.

See the query guide for search tips, the feature list for the full command surface, and vera --help for CLI details.

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 works with vera search, vera grep, and vera references; --timing works with vera search and vera grep. You can place them before or after the subcommand (for example, vera --timing search "auth" or vera references parse_config --raw).

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

If a file is missing from the index and you need the exact reason, run:

vera explain-path path/to/file

Semble benchmark comparison on 1,251 tasks across 63 repositories (Vera v1.2.0 row measured 2026-08-26 on the pre-upgrade Ryzen 7 7600X3D host; Semble column from the 2026-08-23 comparison):

Both tools used the same minishlab/potion-code-16M-v2 embeddings, harness, graded relevance, and suffix-corrected path matching in the scorer. On the 320-task tuning subset, Vera scored 0.8534 versus Semble at 0.8494 nDCG. On the contamination-check independent set, Vera scored 0.7654 versus Semble at 0.7655. See docs/benchmarks.md for the screening tables and historical comparisons.

Latency figures from the 2026-08-26 host are not directly comparable to measurements on the current Ryzen 7 9800X3D host. See the hardware caveat in docs/197-profiling.md and the v1.3.0 notes in docs/whats-new.md. All new ranking knobs added in v1.3.0 ship default off, so the v1.2.0 ranking defaults match v1.3.0 and the quality columns (nDCG, recall, MRR) still represent current behavior.

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

vera agent install installs the Vera skill for supported coding agents and can add a short usage snippet to your project's AGENTS.md, CLAUDE.md, COPILOT.md, or editor rules file.

vera agent install
vera agent install --client all

If you use the skills CLI, you can install Vera there too:

npx skills add VeraTools/Vera

If you skipped the prompt and want to add the instructions manually, use the snippet in the Installation Guide.

See CONTRIBUTING.md.