Vera

Vera

Semantic code search CLI. Combines BM25 keyword matching with vector similarity and cross-encoder reranking to return the most relevant code for a natural-language query.

Workflow

1. Ensure Vera is installed and on PATH (add .vera/ to .gitignore on first use). If missing: references/install.md.
2. Configure exclusions: Vera respects .gitignore by default (no setup needed). If you need custom exclusions, create a .veraignore file (gitignore syntax). Important: .veraignore replaces .gitignore rules entirely. To keep gitignore rules and add your own on top, start .veraignore with #include .gitignore, then add only your extra patterns (do not repeat entries already in .gitignore). Use --exclude for one-off exclusions, --no-ignore to disable all ignore parsing, or --verbose during indexing to see which files are skipped.
3. Index the repo: vera index . (first time) or vera update . (after edits). Use vera index . --verbose to debug exclusion rules.
4. For long sessions, start the watcher: vera watch . (background process, Ctrl-C to stop, 2s debounce). This auto-updates the index on file changes and replaces manual vera update . calls.
5. Get oriented: vera overview returns a project summary: language breakdown, directory structure, entry points, complexity hotspots, and detected conventions (frameworks, patterns, config files). Use this for onboarding before searching.
6. Use vera references <symbol> to find callers; add --callees to see what it calls. vera dead-code lists functions with no callers.
7. Search:

    vera search "authentication middleware"
    vera search "parse_config" --type function --limit 5
    vera search "database connection" --lang rust --path "src/**"
    vera search "OAuth token refresh" "JWT expiry handling" "auth middleware"
    vera search "config" --intent "find where database connection strings are loaded"
    vera search "keybind handling" --scope docs
    vera search "mod loader" --scope runtime --include-generated
    vera search "config loading" --deep    # RAG-fusion query expansion (falls back to iterative symbol-following)
    vera search "auth" --compact            # signatures only: broad exploration in fewer tokens
   

8. Regex search (exact patterns, imports, TODOs). vera grep only searches indexed files, so .veraignore and exclusion rules apply:

    vera grep "fn\s+main"
    vera grep "TODO|FIXME" -i              # case-insensitive
    vera grep "queryClient|invalidateQueries" --path "frontend/src/**"
    vera grep "Authorization" --lang rust --type function
    vera grep "keybind" --scope docs        # scoped to docs
    vera grep "use std::collections" --context 0  # no surrounding lines
    vera grep "handler" --compact           # signatures only
   

9. Use the first results (they are ranked by relevance). Output is markdown codeblocks by default.

Example Output

vera search "hybrid search" --limit 1

```crates/vera-core/src/retrieval/hybrid.rs:58-110 function:search_hybrid
pub async fn search_hybrid(...) -> Result<Vec<SearchResult>> { ... }
```

The info string contains file_path:line_start-line_end and optional symbol_type:symbol_name. Use --json for compact single-line JSON (programmatic consumption). --raw and --timing work with vera search and vera grep, and can appear before or after the subcommand.

Choosing the Right Tool

vera search understands synonyms and related concepts. vera grep matches literal patterns.

Search Scopes

Vera favors source files by default. Use --scope docs for prose and ADRs, --scope runtime for extracted bundles, and --include-generated for minified/dist artifacts.

Search Modes

• Default: full results with code bodies. Best for targeted retrieval ("how does BM25 scoring work?").
• --deep: deep search. Runs a BM25 pre-filter to gather real symbol names and file paths, then feeds them to an LLM to generate query rewrites grounded in actual codebase identifiers. Each rewrite runs a full hybrid search and results fuse with RRF. Requires VERA_COMPLETION_BASE_URL; falls back to iterative symbol-following when unconfigured. Use when initial results miss the mark or you need broader context.
• --compact: signatures only (name, parameters, return type). Fits more results into fewer tokens. Best for broad exploration ("what functions handle auth?"). Works with vera grep too.

Query Strategy

• Describe behavior or intent: "JWT token validation", "request rate limiting", not "code" or "utils".
• Avoid overly broad queries like "authentication" or "tools". Be specific about what aspect you need.
• Match your intent to the query: for documentation, use doc-focused keywords ("setup guide", "configuration README"); for code, use implementation terms ("token refresh logic", "error handling implementation").
• Use 2-3 varied queries to capture different aspects (e.g., "OAuth token refresh", "JWT expiry handling", "auth middleware"). You can pass them in one call: vera search "OAuth token refresh" "JWT expiry handling" "auth middleware".
• Add --intent when the query is ambiguous but your higher-level goal is clear (e.g., vera search "config" --intent "find where database connection strings are loaded from environment variables").
• For known symbol names, search the exact name: vera search "parse_config".
• Start broad, then narrow with --lang, --path, --type, --limit.
• After code changes mid-session, run vera update . before searching again (or use vera watch . to auto-update).

Failure Recovery

• no index found → vera index .
• stale results after edits → vera update .
• local model/ONNX fails → vera doctor --probe, then references/troubleshooting.md
• missing local assets → vera repair
• switch GPU/model backend → vera backend
• API credentials missing → references/install.md
• MCP requested → references/mcp.md

References

• references/install.md: install, setup, API and local config
• references/query-patterns.md: more query examples and rg guidance
• references/troubleshooting.md: common errors and fixes
• references/mcp.md: optional MCP server usage

Source: lemon07r/Vera — distributed by TomeVault.