goloop
An agentic loop on your terminal
Last verified:
What is goloop?
Goloop is an open-source agentic loop CLI that brings AI-powered automation to the command line. It uses a supervisor LLM to plan each step of a task and a worker (Cursor or Claude Code) to execute the actual code edits. State and progress are stored in plain markdown files (.goloop/checkpoint.md) that users can read, diff, and trust, eliminating hidden databases.
Key features include a clean supervisor/worker split where the planner never touches files directly, pluggable backends supporting ChatGPT, OpenAI, and Anthropic for supervision with Cursor CLI or Claude Code for execution, self-extending tools that can author their own supervisor tools, human-in-the-loop capability via ask_user to stop and ask when only the user can answer, and an honest terminal UI with streamed worker reasoning and live iteration logs. The tool is written in Go as a single static binary with no runtime or node_modules dependencies.
Goloop is designed for developers and programmers who want to automate coding tasks, build CLI applications, or iterate on projects with AI assistance. Users point it at a goal (e.g.,
goloop pricing
Pricing model: Freemium
Free and open source under the MIT license. No paid plans mentioned. The tool has no cost to use.
goloop pros
- Open source under MIT license
- Single static binary written in Go
- No runtime or node_modules dependencies
- Simple curl | bash installation
- Clean supervisor/worker separation
- Pluggable backends for multiple providers
- State stored in readable markdown files
- Self-extending tools capability
- Human-in-the-loop ask_user feature
- Honest terminal UI with streamed reasoning
- Live iteration log visualization
- Layered configuration system
- Global auth set once, per-project goals
- Flag overrides for single runs
- Auto-restart after tool authoring
- Checkpoint writes to .goloop/checkpoint.md
- Works on macOS, Linux, and Windows
- Supports ChatGPT, OpenAI, Anthropic
- Worker integrates with Cursor CLI
- Worker integrates with Claude Code
goloop cons
- Requires Go 1.24+ for go install method
- Only supports Cursor or Claude Code as workers
- Limited to ChatGPT, OpenAI, Anthropic supervisors
- No Windows native install script shown
- Project config.yaml required to run
- No GUI interface, CLI only
- May hallucinate without ask_user trigger
- No built-in test running capability mentioned
- Depends on external API provider availability
- No concurrent task execution mentioned
Frequently asked questions about goloop
What is Goloop?
Goloop is an open-source agentic loop CLI that runs on the command line. A supervisor LLM plans every iteration and a worker (Cursor or Claude Code) does the actual code edits. State lives in plain markdown files you can read, diff, and trust.
How do I install Goloop?
You can install Goloop by running: curl -fsSL https://raw.githubusercontent.com/mantyx-io/goloop/main/scripts/install.sh | bash for macOS/Linux/WSL. Alternatively, you can install with Go using: go install github.com/mantyx-io/goloop/cmd/goloop@latest
How do I configure Goloop?
Run goloop configure once to pick your provider, models, and sign in. Your settings are saved to ~/.goloop/config.yaml. Then run goloop login to complete authentication.
What supervisors does Goloop support?
Goloop supports ChatGPT subscription, OpenAI API, and Anthropic API as supervisor backends. You can swap between them freely during configuration.
What workers does Goloop support?
Goloop supports Cursor CLI and Claude Code as workers that perform the actual code edits in your repository.
Where does Goloop store state?
State lives in plain markdown files, not a hidden database. Progress, blockers, and the iteration log are stored as markdown files you can diff and edit. Checkpoints are written to .goloop/checkpoint.md.
What is the supervisor/worker split?
The planner (supervisor) never touches files directly. A dedicated worker does every edit. This clean separation ensures predictable runs and keeps planning separate from execution.
How does human-in-the-loop work?
When only you can answer a question, Goloop stops and asks using the ask_user action instead of hallucinating a decision. This prevents the AI from making incorrect assumptions.
What configuration files does Goloop use?
Goloop uses two config files: global ~/.goloop/config.yaml for provider, models, worker, and loop defaults (set once), and project <dir>/.goloop/config.yaml for the objective specific to that repository (required to run). Run flags can override either for a single invocation.