# claw-code-free

**Repository Path**: lr998/claw-code-free

## Basic Information

- **Project Name**: claw-code-free
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-04-05
- **Last Updated**: 2026-04-05

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# Claw Code — Free Local LLM Edition

An agentic coding CLI that works with **free local models** via [Ollama](https://ollama.com/), no API keys required.

Cloned from [ultraworkers/claw-code](https://github.com/ultraworkers/claw-code) with modifications to support OpenAI-compatible providers (Ollama, LM Studio, etc.), so you can use free models like Qwen, Llama, Mistral and more.

## What We Changed

The original project only supported Anthropic's Claude API. We modified the CLI to use **dynamic provider detection** — when you pass a non-Anthropic model via `--model`, it automatically routes to your local Ollama instance through the OpenAI-compatible API.

**Key changes:**
- `rust/crates/claw-cli/src/main.rs` — Switched from hardcoded `ClawApiClient` to `ProviderClient` with model-based provider routing
- `rust/crates/api/src/client.rs` — Added base URL support for the default auth path

## Quickstart

### Prerequisites

1. Install [Rust](https://rustup.rs/)
2. Install [Ollama](https://ollama.com/)
3. Pull a model:
   ```bash
   ollama pull qwen2.5-coder:7b
   ```

### Build

```bash
cd rust
cargo build --release
```

### Run

**PowerShell:**
```powershell
$env:ANTHROPIC_API_KEY = ""
$env:ANTHROPIC_AUTH_TOKEN = ""
$env:OPENAI_API_KEY = "dummy"
$env:OPENAI_BASE_URL = "http://localhost:11434/v1"

.\target\release\claw.exe --model qwen2.5-coder:7b
```

**Bash / macOS / Linux:**
```bash
export ANTHROPIC_API_KEY=""
export ANTHROPIC_AUTH_TOKEN=""
export OPENAI_API_KEY="dummy"
export OPENAI_BASE_URL="http://localhost:11434/v1"

./target/release/claw --model qwen2.5-coder:7b
```

> Make sure Ollama is running first (`ollama serve` or open the Ollama desktop app).

### Recommended Models

| Model | Size | Good at | Command |
|-------|------|---------|---------|
| `qwen2.5-coder:7b` | ~4GB | Coding | `ollama pull qwen2.5-coder:7b` |
| `qwen2.5-coder:14b` | ~9GB | Coding (better) | `ollama pull qwen2.5-coder:14b` |
| `mistral:7b` | ~4GB | General + tool calling | `ollama pull mistral:7b` |
| `llama3.1:8b` | ~5GB | General purpose | `ollama pull llama3.1:8b` |
| `llama3.1:70b` | ~40GB | Best quality (needs lots of RAM) | `ollama pull llama3.1:70b` |

### Known Limitations

- Small local models (7B-14B) may **not reliably use tools** (file read/write, bash, etc.). They tend to describe what to do rather than actually doing it. For full agentic functionality, larger models (70B+) or cloud APIs work better.
- Tool/function calling quality varies by model. Mistral tends to handle it better than others at small sizes.

## Original Project

This project is based on [ultraworkers/claw-code](https://github.com/ultraworkers/claw-code), which is a Rust port of the Claw Code agent harness. All credit for the original implementation goes to the upstream authors.

## Disclaimer

- This repository does **not** claim ownership of the original Claw Code source material.
- This repository is **not affiliated with, endorsed by, or maintained by the original authors**.