rune-code 0.15.0

A terminal first, opinionated AI coding agent written in Python. Works with almost any model via Pydantic-AI.

Rune Code

Rune's a terminal first coding agent, it's my personal forever ongoing personal project.

It builds on the excellent Pydantic-AI which among many things lets you easily switch between different models (Claude vs Gemini) as well as providers (Claude on Bedrock, Gemini on Vertex AI, etc.)

---

Table of Contents

1. Features
2. Prerequisites
3. Installation & Configuration
4. Quick Start: Your First Session
5. Usage Cheatsheet
6. Core Capabilities
7. For Developers
8. License

---

Features

Capability	Description
Interactive Chat	Multi-line editing, history, and command completion via prompt-toolkit.
Advanced Tooling	Edit files with fuzzy diffs, grep the codebase, run shell commands, and execute Python in a persistent Jupyter kernel.
Rich TUI	A clean, colorful terminal UI powered by Rich, with syntax highlighting, tables, and live-streaming output.
Session Management	Conversations and session context are automatically saved, allowing you to resume a session exactly where you left off. For example todos are automatically saved and loaded along with messages.
Task Planning	Rune can manage a TODO list to break down complex tasks, track progress, and ensure it completes all steps.
Extensible	Easily add new tools by dropping a function into the src/rune/tools/ directory.

---

Prerequisites

• Python 3.10 or higher.
• An API key for a supported LLM provider (e.g., OpenAI, Google, Anthropic).

---

Installation & Configuration

1. Installation

Install Rune - I'd recommend uv nowadays.

# Recommended
uv add rune-code

# Alternatively, using pip
pip install rune-code

2. Configuration

Rune builds on top of Pydantic-AI - make sure to setup auth/config correctly. See here.

# Example for OpenAI
export OPENAI_API_KEY="sk-..."

# Example for Google
export GOOGLE_API_KEY="..."

You can also specify which model to use. If not set, Rune defaults to a safe but capable model. Model list here.

# Optional: Specify a model
export RUNE_MODEL="google-gla:gemini-2.5-pro"

# Google Vertex Example

export RUNE_MODEL="google-vertex:gemini-2.5-pro"

# OpenAI Example
export RUNE_MODEL="openai:gpt-4"

# Anthropic Example
export RUNE_MODEL="anthropic:claude-sonnet-4-20250514"

# Bedrock Example
export RUNE_MODEL="bedrock:us.anthropic.claude-sonnet-4-20250514-v1:0"

Within the chat you can also use the slash command /model <model_name> to switch models. It'll allow you to tab complete the model name.

---

Quick Start: Your First Session

Let's run a simple task to see Rune in action: listing the files in the current directory.

Step 1: Start Rune

Run the rune command in your terminal.

rune

You will be prompted to start a new session.

Step 2: Ask Rune to List Files

At the prompt, ask Rune to list the files.

> Use the list_files tool to show me the files in the current directory.

Step 3: See the Result

Rune will execute the list_files tool and display the output in a clean, tree-like format.

This simple interaction demonstrates the core loop: you give Rune a goal, and it uses its tools to accomplish it.

---

Usage Cheatsheet

Action	Command	Notes
Start a new chat	rune	Choose "Start new session" from the menu.
Resume a session	rune	Pick a recent session from the list.
Exit the chat	/exit or Ctrl-D
Interrupt a task	Ctrl-C	Stops the current operation.
Save a snapshot	/save [name]	Saves the current session state to .rune/snapshots/.
Change model	/model <name>	e.g., /model google:gemini-1.5-pro. Use Tab to complete.
List models	rune models list	Lists all supported models grouped by provider.
Change directory	run_command("cd path/to/dir")	Changes the agent's working directory for tool use.

---

Core Capabilities

Rune's power comes from its built-in tools. Here are a few examples of what you can ask it to do.

• Grep the codebase:

  "Find all occurrences of the run_command function."

• Edit a file with a precise diff:

  "In src/rune/main.py, find the run_agent_turn function and add a print statement at the beginning that says 'Starting turn'."

• Run a shell command and analyze its output:

  "Run ls -l and tell me which file was modified most recently."

• Manage a task list for a complex change:

  "I need to add a new feature. First, create a new file called features.py. Second, add a function to it called new_feature. Finally, add a test for it in tests/test_features.py."

---

For Developers

The following sections are for those interested in contributing to or learning about the architecture of Rune.

Architecture Overview

flowchart TD
    subgraph CLI
        A["chat.py
        prompt-toolkit"]
    end
    subgraph Agent
        B["pydantic_ai.Agent"] --> C["rich_tool wrapper"]
    end
    subgraph Tools
        D["edit_file
        run_python
        grep
        …"]
    end
    subgraph UI
        E["render.py
        Rich console
        LiveDisplayManager"]
    end
    subgraph Persistence
        F[".rune/sessions/*.json"]
    end

    A -->|prompt| B
    B -->|XML tool call| C
    C -->|exec| D
    D -->|ToolResult| C
    C -->|renderable| E
    B -->|assistant text| E
    A <--> F

Core Libraries

Domain	Library
LLM Orchestration	pydantic-ai
Terminal UI	Rich
CLI Framework	Typer + prompt-toolkit
Diff/Patch Engine	Custom DiffApplyer + difflib
HTTP & Markdown	httpx, html-to-markdown
Interactive Python	jupyter_client, ipykernel
Git-style Ignores	pathspec

Testing & Linting

We use pytest for testing, ruff for linting and formatting, and mypy for type checking.

# Run all unit tests
pytest

# Check for linting errors and format code
ruff check .
ruff format .

# Run static type checking
mypy src

Contributing

Contributions are welcome! Please follow these steps:

1. Fork the repository and create a feature branch.
2. Follow the style conventions enforced by ruff.
3. Add unit tests for any new functionality.
4. Ensure all checks (pytest, ruff, mypy) pass.
5. Submit a pull request.

---

License

Rune is licensed under the Apache 2.0 License.