# agent0

A local AI coding assistant that runs in your terminal. You type questions or instructions, and the AI reads your code, edits files, runs commands, and searches the web — all from a simple text interface.

No cloud accounts or API keys required for local use. Your code stays on your machine.

## What it does

agent0 gives you an AI assistant that understands your codebase. You can ask it to:

- **Read and edit files** — it can open any file in your project, make targeted edits, or create new files
- **Run shell commands** — compile code, run scripts, install packages, anything you'd do in a terminal
- **Search your code** — find files by name patterns or search inside files for specific text
- **Look up Clojure docs** — instant documentation for Clojure standard library functions
- **Research the web** — search the internet and read web pages to answer questions
- **Run skills** — execute predefined workflows you set up for common tasks

Everything happens in a terminal UI with a chat-style interface. You type, the AI responds and uses tools as needed, and you see what it's doing in real time.

## Requirements

### Self-hosted (local, no API keys)

This is the default setup. The AI model runs entirely on your machine.

- **[Babashka](https://github.com/babashka/babashka)** — a fast Clojure scripting runtime (used to run agent0)
- **[Ollama](https://ollama.com)** — runs AI models locally on your computer
- **A downloaded model** — agent0 defaults to `qwen3-coder-next`, a coding-focused model

Hardware considerations: local models need a decent GPU or a machine with enough RAM. Smaller models work on modest hardware; larger models need more resources. Check [Ollama's model library](https://ollama.com/library) for size and requirements.

### Remote LLM (Ajet Cloud, OpenAI, Anthropic)

If you'd rather use a cloud-hosted model instead of running one locally, agent0 can connect to any remote server that speaks the Ollama API:

- **[Babashka](https://github.com/babashka/babashka)** — still required to run agent0 itself
- **A remote Ollama-compatible endpoint** — set `OLLAMA_HOST` to the server URL

Some providers that work:
  - [Ajet Cloud](https://ajet.fyi) — managed Ollama-compatible hosting
  - Any remote machine running Ollama
  - OpenAI or Anthropic via an Ollama-compatible proxy (e.g. [LiteLLM](https://github.com/BerriAI/litellm))

With a remote provider, you don't need Ollama installed locally or a powerful machine — the model runs on the remote server.

## Installation

### 1. Install Babashka

**macOS:**
```bash
brew install borkdude/brew/babashka
```

**Linux:**
```bash
bash < <(curl -s https://raw.githubusercontent.com/babashka/babashka/master/install)
```

See the [Babashka install docs](https://github.com/babashka/babashka#installation) for other methods.

### 2. Set up a model

**For local use** — install Ollama and pull a model:
```bash
# Install Ollama from https://ollama.com, then:
ollama pull qwen3-coder-next
```

**For remote use** — point to your provider:
```bash
export OLLAMA_HOST="https://your-provider-url"
```

### 3. Clone the repository

```bash
cd ~/repos  # or wherever you keep projects
git clone https://git.ajet.fyi/ajet/agent0.git
```

Dependencies (including the TUI framework) are fetched automatically via git deps.

### 4. Add to your PATH

Make the `agent` command available from anywhere:

```bash
# Add to your shell profile (~/.bashrc, ~/.zshrc, etc.)
export PATH="$HOME/repos/agent0:$PATH"
```

Then restart your terminal or run `source ~/.bashrc`.

## Usage

```bash
agent                          # Start a new conversation
agent "explain this codebase"  # Start with a question or instruction
agent --continue               # Resume your most recent session
agent --session <id>           # Resume a specific session by ID
```

Once running, type your message and press Enter. The AI will respond, using tools as needed. You'll see what it's doing — reading files, running commands, searching — in real time.

## Configuration

### Environment variables

| Variable | What it does | Default |
|---|---|---|
| `AGENT_MODEL` | Which AI model to use | `qwen3-coder-next` |
| `OLLAMA_HOST` | Where the model is running | `http://localhost:11434` |

### Project context

agent0 automatically reads instructions from your project to give the AI relevant context. It checks for these files (in order):

1. `.agent0/context.md` — agent0-specific project instructions
2. `CLAUDE.md` — also used by Claude Code
3. `.cursorrules` — also used by Cursor
4. `.github/copilot-instructions.md` — also used by GitHub Copilot

You can also put global instructions (applied to all projects) in `~/.config/agent0/context.md`.

### Skills

Skills are reusable prompt templates you trigger with `/name`. Define them in:

- `~/.config/agent0/skills.md` — available in all projects
- `.agent0/skills.md` — project-specific skills

Example skill definition:
```markdown
# /deploy <env>

Deploy the application to the {env} environment. Run the deploy script,
verify it completes successfully, and report the result.
```

Then use it: type `/deploy staging` and it expands into the full prompt.

## Sessions

Every conversation is automatically saved. You can pick up where you left off:

```bash
agent --continue               # Resume the latest session
agent --session <id>           # Resume a specific session
```

Sessions are stored in `~/.local/share/agent0/sessions/`. Logs are in `~/.local/share/agent0/logs/`.