Kolega Code
Kolega Code is a local-first terminal coding agent that lets the model write its own multi-agent workflows via the Gigacode engine. It solves the problem of tasks too broad for a single agent loop by having the model author a Python orchestration program that the runtime executes. It is for developers who want an open, provider-agnostic alternative to tools like Claude Code for complex coding tasks.
✨ Key features
- Gigacode engine: model writes Python orchestration for multi-agent workflows
- Plan/Build modes with read-only investigation and reviewable task lists
- Terminal execution with streamed output and permission controls
- Web search and browsing with DuckDuckGo default, plus configurable backends
- MCP server support (streamable_http, sse, stdio) as permission-gated tools
- Editor integration via Agent Client Protocol for Zed and VS Code
🎯 Use cases
- Repo-wide code review with parallel sub-agents
- Large-scale migrations with independent workstreams
- Complex plans requiring pipelines, judge panels, and synthesis gates
- Automated scripting and CI with one-shot
askcommand - Interactive day-to-day development with plan/build modes
📦 Installation
🧰 Requirements: Python 3.11+, an API key or ChatGPT sign-in for at least one supported model provider, and a terminal that supports a modern TUI.
curl -fsSL https://kolega.dev/install-kolega-code.sh | sh
Or with uv (or pip):
uv tool install kolega-code
# or: pip install kolega-code
Verify the install:
kolega-code --version
🚀 Usage
Start a session in your project:
kolega-code .
Then connect a model via the first-run wizard. Use Shift+Tab to switch between Plan and Build mode, or run /gigacode on for broad tasks.
For one-shot use:
kolega-code ask "summarize this repository" --project .
⚠️ Good to know
Gigacode is off by default and must be enabled with /gigacode on; the README also notes that mechanics, guarantees, and limitations are documented in how-gigacode-works.md, but no specific limitations are listed.
❓ FAQ
How does Gigacode work?
Gigacode lets the model write a Python program that orchestrates multiple sub-agents using a runtime API (agent, parallel, pipeline, phase, budget). The runtime executes the program, enforcing concurrency caps and token budgets.
Can I resume an interrupted run?
Yes, the runtime journals each agent call to disk. Use resume_from_run_id to replay unchanged calls from the journal at zero token cost, even if the script was edited or reordered.
Which model providers are supported?
Anthropic, OpenAI (API or ChatGPT sign-in), Google, Groq, Together.ai, Fireworks.ai, xAI/Grok, DashScope/Qwen, OpenRouter, Moonshot/Kimi, DeepSeek, Z.AI/GLM, Kimi Coding Plan, Ollama Cloud, and local Llama.
How do I enable Gigacode?
Gigacode is off by default. Enable it in a session with the /gigacode on command.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.