VT Code is a Rust-based terminal coding agent with semantic code intelligence (Tree-sitter parsers for Rust, Python, JS/TS, Go, Java) and ast-grep for structural search/refactors. It runs as a fast CLI/TUI and speaks to multiple LLM providers with automatic failover and token-efficient context handling. MIT-licensed.
I wanted a coding agent that’s editor-agnostic, reproducible via a single vtcode.toml, and optimized for long sessions without burning tokens. Rust for performance/reliability; Tree-sitter/ast-grep for precise code ops.
Semantic code tools (Tree-sitter) + ast-grep refactors built-in. Multi-provider: OpenAI, Anthropic, xAI, DeepSeek, Gemini, OpenRouter, Z.AI; also Ollama for local models. Zed ACP integration so it works natively inside Zed as an external agent.
Works as an Agent Client Protocol client; supports file context, tool calls, JSON-RPC over stdio. Setup steps + example settings are in the repo docs.
This is early but usable. I’m iterating on context curation, component extraction, and safer tool policies. Feedback on failure modes + DX is welcome.
I wanted a coding agent that’s editor-agnostic, reproducible via a single vtcode.toml, and optimized for long sessions without burning tokens. Rust for performance/reliability; Tree-sitter/ast-grep for precise code ops.
Semantic code tools (Tree-sitter) + ast-grep refactors built-in. Multi-provider: OpenAI, Anthropic, xAI, DeepSeek, Gemini, OpenRouter, Z.AI; also Ollama for local models. Zed ACP integration so it works natively inside Zed as an external agent.
Works as an Agent Client Protocol client; supports file context, tool calls, JSON-RPC over stdio. Setup steps + example settings are in the repo docs.
This is early but usable. I’m iterating on context curation, component extraction, and safer tool policies. Feedback on failure modes + DX is welcome.