ragent

A local-only agentic AI framework written in Rust.

Run your own LLM (via Ollama), execute tools autonomously, and integrate with code editors like ZED and VS Code — all without sending data to the cloud.

Goals

• Full local control: Swap the underlying LLM, inspect every request/response, keep all data on your network
• Agentic tool use: Autonomous multi-step tool execution (file ops, shell, web research) driven by a local model
• Editor integration: OpenAI-compatible API server so ZED, VS Code, or any compatible client can use ragent as an LLM provider
• Debuggability: Structured logging of the entire agent loop — see exactly what the model asked for and what tools returned
• Extensible skills: Define specialised agent profiles (system prompt + tool set) per task type

Status

🚧 Early development — see PLAN.md for the full roadmap.

Prerequisites

• Rust (stable toolchain)
• Ollama running locally with at least one model pulled (e.g. ollama pull llama3.2)

Quick Start

# Clone
git clone https://github.com/<you>/ragent.git && cd ragent

# Build
cargo build

# Run (once Phase 1 is complete)
cargo run -- --model llama3.2

Licence

GPL-3.0 — see LICENSE.md.