tpt-anvil

Rust

Local-first AI coding assistant. Features a Rust daemon, Tree-sitter + BM25/vector indexing, GGUF inference, and VS Code/JetBrains extensions.

0 stars0 forks0 watchers30 open issues

Languages

Rust78.7%TypeScript12.1%Kotlin8.8%JavaScript0.4%
README

TPT Anvil

The open-source, privacy-first AI development environment.

Status: pre-alpha — active development

Anvil is a fully open-source alternative to GitHub Copilot and Cursor. Everything runs on your machine — your code never leaves.

Features

  • Local LLM inference — llama.cpp, candle (pure Rust), or Ollama; any GGUF model
  • GPU acceleration — CUDA, ROCm, WebGPU
  • Codebase-aware context — Tree-sitter AST + hybrid BM25/vector semantic search
  • Cloud fallback — OpenAI, Anthropic, OpenRouter, or any OpenAI-compatible endpoint
  • VS Code extension — chat panel, slash commands, diff-based edits
  • JetBrains plugin — IntelliJ IDEA, PyCharm, GoLand, and all JetBrains IDEs
  • Privacy-first — 100% local by default, zero telemetry

Slash Commands

CommandDescription
/generateGenerate code from a description
/testGenerate unit tests for selected code
/explainExplain selected code in plain language
/fixDiagnose and fix selected code
/docsGenerate docstrings and documentation

Architecture

┌─────────────────────────────────────┐
│  IDE Extension (VS Code / JetBrains)│
│  TypeScript / Kotlin                │
└────────────────┬────────────────────┘
                 │ JSON-RPC (Unix socket / named pipe)
┌────────────────▼────────────────────┐
│         anvil-daemon (Rust)         │
│  ┌─────────────┐ ┌───────────────┐  │
│  │  Inference  │ │   Indexer     │  │
│  │  Backends   │ │  Tree-sitter  │  │
│  │  llama.cpp  │ │  BM25+Vector  │  │
│  │  candle     │ │  sqlite-vec   │  │
│  │  Ollama     │ └───────────────┘  │
│  │  Cloud APIs │ ┌───────────────┐  │
│  └─────────────┘ │  Capabilities │  │
│                  │  /generate    │  │
│                  │  /test /fix   │  │
│                  └───────────────┘  │
└─────────────────────────────────────┘

Getting Started

Installation instructions coming soon. See docs/getting-started.md.

License

Dual-licensed under MIT OR Apache-2.0 — your choice.

Copyright (c) 2026 TPT Solutions