The limits of text search
Text search misses structure
ripgrep finds every line containing authenticate — but can't tell you who calls it, what it returns, or whether it's used at all.
Embeddings are slow and imprecise
Semantic search via embeddings takes 1,400ms+ and returns false positives. ~60% precision means you still read code that doesn't match.
Cloud tools send your code offsite
Sourcegraph, GitHub Copilot, and similar tools require uploading your codebase. Not an option for private or regulated code.
STRUCTURAL QUERIES
Exact answers.
Not guesses.
sqry parses your code like a compiler — building a graph of every symbol, call, import, and type relationship. Queries return exact structural results in milliseconds.
MCP INTEGRATION
AI assistants that query,
not guess.
Connect Claude, Codex, or Gemini to sqry via MCP. Your AI assistant gets exact structural answers instead of probabilistic guesses.
$ sqry mcp setupsemantic_search
Search symbols by name, kind, language, and pathdirect_callers
Exact list of every caller of a functioncall_hierarchy
Full inbound and outbound call treefind_unused
Dead code detection across the codebasesemantic_diff
Structural changes between git commitsWorks in your editor.
sqry ships an LSP server with standard protocol support plus 27 custom methods.
sqry lsp --stdio35 languages supported
TIER 1 — Full relation tracking (28 languages)
TIER 2 — Symbols + imports (7 languages)
Built for speed
LINUX KERNEL BENCHMARK
11 million nodes.
Under 2 minutes.
sqry indexes the entire Linux kernel — 28M lines of C across 63,000 files — in under 2 minutes. The resulting graph has 11.2M nodes and 18.3M edges. Query it instantly.
NATURAL LANGUAGE
Ask in plain English.
sqry ask translates natural language into safe, validated sqry commands. A compact ML classifier (22M params, runs locally) achieves 99.75% intent accuracy with sub-3ms latency.
- ✓ 4-tier confidence system (execute / confirm / disambiguate / reject)
- ✓ Whitelist-only command generation — no shell injection
- ✓ Unicode homoglyph detection
- ✓ Works without ML model via rule-based fallback
- ✓ Available as CLI, MCP tool, and LSP endpoint
SECURITY & SUPPLY CHAIN
Built to be trusted.
Every release is signed, attested, and reproducible. We ship SBOMs with every binary, fuzz the parser continuously, and audit the entire dependency tree on every commit.
Signed & attested builds
Every binary is signed with Sigstore keyless signing and ships with per-platform SLSA provenance. No long-lived keys — attestation flows through GitHub OIDC.
SBOM & vulnerability exchange
Every release ships CycloneDX and SPDX SBOMs, Grype vulnerability scans, and OpenVEX documents. Full dependency transparency for your compliance workflows.
Audited dependency tree
Every dependency is audited via cargo-vet with imports from Mozilla, Google, and Bytecode Alliance. License policy and CVE checks run on every commit.
Continuous fuzzing & UB detection
The query parser is fuzzed on every PR and in extended nightly runs with AddressSanitizer. Miri checks for undefined behavior weekly. Malformed input tests cover all 34 language parsers.
Mutation testing & coverage gates
Weekly mutation testing validates that tests catch real bugs, not just pass. LLVM-based code coverage with enforced thresholds. Clippy with zero-warning policy across all platforms.
Unsafe code tracking
Weekly cargo-geiger audits track every unsafe block across all crates. Quarterly security audits combine RustSec advisory checks, license enforcement, and supply chain verification.
sqry is open source.
Support subscriptions fund continued development, priority bug fixes, and enterprise support.