The Agentic TUI for Go codebases ;) (github.com)

🤖 AI Summary
Vogte is an agentic terminal UI built specifically for Go repositories that combines AST-driven context extraction with LLMs and native Go tooling to help developers inspect, patch, and validate code. It compresses repo context by extracting structs, interfaces, methods and signatures (rather than dumping whole files), asks the model which files it needs, then supplies those files so the LLM can generate fixes or new code. Vogte can run in an “Agent” mode that applies patches directly (currently line-based, with AST-based patching explored), and it immediately runs a “Sanity Check” (e.g., go vet ./...) to give instant project health feedback. Significance for the AI/ML and dev-experience communities lies in its language-specific approach: AST parsing reduces prompt size and noise, Go toolchain integration (tests, go vet, goimports) gives fast, reliable validation, and the two-step file-retrieval flow minimizes unnecessary context exposure and API cost. It’s CLI-first, near-zero config (auto-uses OPENAI_API_KEY or ANTHROPIC_API_KEY; can auto-select GPT-5), supports multiple LLMs (tested with GPT-4 and Claude Sonnet), and outputs a compressed repository context for reuse. Limitations include rough edges in agentic patching and no agentic retry loop on failure, but Vogte represents a practical pattern for tighter LLM + language-tooling integration in real codebases.
Loading comments...
loading comments...