Open Context Engine

Local-first codebase indexing, and an AI agent grounded in your code.

Open Context Engine indexes your workspace into a SQLite store on your machine: AST-aware chunks, a code graph of imports, calls and inheritance, keyword (BM25) search, and — with an embedding provider — semantic vectors. The chat agent searches that index as it works, so its answers come from the code you actually have.

Features

Chat grounded in your index. A sidebar chat with streaming markdown and code-block copy, insert and apply, backed by ranked retrieval from your workspace.
Agent edits you can review. Each changed file gets a native diff, undo and redo, and every turn gets an "undo all". The agent's shell tool is off by default.
Search from anywhere. Quick Search (Ctrl+Shift+K, or Cmd+Shift+K on macOS), Find Similar Code, and Explain with Context from the editor.
A live index. Files re-index on save. The Index Health panel shows what is indexed, and Debug Retrieval shows why a query ranked what it did.
AST-aware chunking for 13 languages: TypeScript, JavaScript, Python, Go, Rust, Java, C#, C, C++, Ruby, PHP, Kotlin and Swift. Other text files are chunked by lines.

Getting started

Open a folder. The extension indexes it on startup.
With no embedding key, search runs on keyword ranking and works straight away.
For semantic search, run Open Context: Set Embedding API Key with a Voyage key — Voyage (voyage-code-3) is the default provider. To use OpenAI instead, set openContext.embedding.provider to openai first; the key is stored for whichever provider is selected. For free local embeddings, run ollama pull nomic-embed-text and set openContext.embedding.provider to ollama — no key needed. Each provider uses its own default model unless you set openContext.embedding.model.
For chat, pick openContext.llm.provider — OpenAI, Anthropic, Google, Ollama, or any OpenAI-compatible endpoint via custom — and run Open Context: Set LLM API Key, or choose both from the chat's model picker.

Keys are kept in VS Code's SecretStorage.

Requirements

VS Code 1.103 or newer.
Windows x64, macOS (Intel and Apple Silicon), or Linux x64/arm64 with glibc 2.34 or newer (Ubuntu 22.04, Debian 12, RHEL 9 and later). Remote hosts — SSH, WSL, Dev Containers, Codespaces — work on the same platforms. Alpine (musl) is not supported.

Privacy

Indexing, the store, and keyword search run on your machine. Code leaves it only for the embedding and LLM providers you configure; use Ollama for both to stay fully offline. There is no telemetry.

Settings

Setting	Default
`openContext.embedding.provider`	`voyage`	`voyage`, `openai`, or `ollama`
`openContext.llm.provider`	`openai`	`openai`, `anthropic`, `google`, `ollama`, or `custom`
`openContext.agent.allowEdits`	`true`	Lets the agent edit, create and delete files
`openContext.agent.shell.enabled`	`false`	Lets the agent run shell commands in the workspace
`openContext.autoIndex`	`true`	Re-indexes files as they change
`openContext.indexOnStartup`	`true`	Indexes the workspace when VS Code starts

The same engine elsewhere

Open Context Engine also runs as a CLI and an MCP server for Claude Code, Cursor and other MCP clients (npm install -g open-context-engine), and as Trace, an agent workspace for the terminal and the browser. See opencontextengine.com.

License

MIT. Paid plans add team features; see pricing. Issues and source: github.com/Dhaxor/open-context-engine.