Gortex indexes source code into a persistent knowledge graph of functions, classes, call chains, HTTP routes, and other relationships, and exposes it to AI coding agents and IDEs through a CLI, an MCP server, an HTTP API, and a web UI. Instead of reading whole files, an agent asks the graph for the symbols, callers, and dependencies it needs, which cuts the tokens spent on context. A savings report tracks tokens saved against naive file reads.
Parsing covers 257 languages and grammars through tree-sitter, regex, and signature tiers, with compiler-grade resolution via language servers for Python, TypeScript, Go, Java, Rust, C and C++, and others. Several repositories share one graph by default, so references and call chains cross repository boundaries, and Gortex matches API contracts (HTTP routes, gRPC, GraphQL, message topics, WebSocket events, OpenAPI specs, and Temporal workflows) between providers and consumers. A precomputed reach index answers blast-radius questions such as what breaks if a function changes.
Gortex ships as a single static binary for macOS, Linux, and Windows that runs as a long-lived daemon. One daemon serves every editor window and MCP client, watches the file system for changes, and stores the graph in SQLite. Its setup step configures the coding agents it detects on the machine, and 19 are supported, including Claude Code, Cline, Continue, OpenCode, Codex CLI, Aider, Kilo Code, OpenClaw, Hermes, and Oh My Pi.
Gortex has been measured indexing large repositories on an Apple Silicon laptop:
| Repository | Files | Nodes | Edges | Index time |
|---|---|---|---|---|
| torvalds/linux | 70,333 | 1,690,174 | 6,239,570 | about 3 min |
| microsoft/vscode | 10,762 | 204,501 | 808,902 | about 1 min |
Features
- Code graph: functions, classes, call chains, and routes indexed into a persistent, provenance-tiered graph
- Language coverage: 257 languages and grammars, plus Jupyter and Databricks notebooks
- Cross-repository graph: references, call chains, and impact analysis across many repositories with per-session isolation
- API contract matching: pairs providers and consumers for HTTP, gRPC, GraphQL, message topics, WebSocket, OpenAPI, and Temporal
- MCP tools: more than 100 tools for symbol lookup, call chains, blast radius, dataflow, clone detection, and refactoring
- Semantic search: built-in embeddings with full-text and vector fusion, with optional MiniLM, Ollama, or OpenAI embeddings
- Speculative execution: previews what an edit would change without touching disk, plus shadow graphs of unsaved editor buffers
- Safety checks: flags broken callers, guard violations, and stale docs before changes ship
- PR review: triages open pull requests by blast radius and emits line-anchored findings with a verdict
- Agent setup: configures each detected coding agent, with 19 supported
- Web UI and HTTP API: a versioned API, Streamable HTTP MCP transport, and a 3D graph visualization
- Optional LLM features: llama.cpp, Anthropic, OpenAI, Ollama, Claude Code or Codex CLI, Gemini, Bedrock, DeepSeek, and Requesty
