Open source · GitHub Issue to release · updated 2026-10-20

An open source AI coding agent, from Issue to release

Q: How do I pause new tasks?

On Linux, systemctl --user disable --now orbi@1.timer pauses future starts for the default instance; an already running task continues. Use your actual timer names, and pause every enabled instance. macOS uses launchd instead of systemctl. See Operations for the Linux timer and running-service checks.

Orbi is an open source AI coding agent for GitHub repositories. Use Orbi Cloud to skip local setup and runtime maintenance, or run Orbi on your own machine. Describe a change and its acceptance criteria in a GitHub Issue. Orbi implements the change and opens a pull request (PR) for independent review. It merges once the checks and required approvals pass. Releases follow your repository’s rules through a separate Issue.

Try Orbi Cloud free →Self-host Orbi ↗

BEFORE YOU CONNECT

Pick an open source coding agent

Use Orbi when a GitHub Issue is a useful handoff: a clear task goes through implementation, independent review and merge.

Put the task and acceptance criteria in an Issue.
Describe the result and how to verify it. You can keep an interactive coding agent for edits you want to steer in a terminal or IDE; Orbi gives you an Issue-based delivery workflow alongside it.

Decide who owns merge and release.
Orbi merges reviewed changes once repository checks and required approvals pass. To approve each merge yourself, require approval through GitHub branch protection; Orbi waits for it. Request a release in a separate Issue with a target version and both ai-ready and ai-release labels. Related tasks and release checks must be complete before it creates a tag and GitHub Release.

Choose who runs and maintains it.
Orbi Cloud hosts the runtime. With self-hosting, you operate the runner and configure its model provider and GitHub credentials. A hosted model can still receive code context; a local model changes the inference location, while Issues and PRs still live on GitHub. The provider guide explains the supported connections. If you prefer hosted operation, compare self-hosted and Cloud below.

For a broader shortlist, read OpenCode alternatives and Orbi vs OpenHands. Compare the configured workflow, too: OpenCode can also run in GitHub Actions, so check each tool’s GitHub integration as well as its terminal interface.

Start with one small Issue whose result you can verify. Check the PR and its review evidence before handing over a larger backlog.

TWO WAYS TO RUN IT

Use Orbi Cloud or self-host?

Use Orbi Cloud to skip installation and maintenance. Self-host if you want to choose and manage your own environment and model. Both start with GitHub Issues and follow your repository’s rules for coding, independent review and merge.

	Self-hosted	Orbi Cloud
Getting started	Prepare a machine, install Orbi, connect a model and repository, and configure scheduled runs.	Sign in with GitHub, authorize the Orbi App and connect a repository. No local installation.
Maintenance	You maintain the machine, Orbi updates and model connection.	Orbi deploys and maintains the hosted runtime.
Cost	No Orbi software fee. You cover the machine and model costs. The cost report explains our model-cost estimates.	A monthly or annual subscription covers hosted operation and the model usage included in your plan.
Usage limits	Orbi sets no self-hosted delivery quota. Machine resources and your model service still have their own limits.	Paid plans have a monthly model-usage allowance. See pricing for the included usage.

Comparing managed agents instead? See Orbi vs Devin for setup, pricing and delivery differences.

SELF-HOSTED SETUP

Self-host an open source AI coding agent

You need a machine that can keep Orbi running, a model connection and GitHub access. Use a systemd user session on Linux or launchd on macOS. The getting-started guide covers repository selection, configuration and service setup; it also records the current limits of macOS verification.

curl -LsSf https://raw.githubusercontent.com/orbi-build/orbi/main/install.sh | bash

Prepare git, curl and an authenticated GitHub CLI (gh ≥ 2.94). The installer clones over SSH, so GitHub SSH access must work too. If Pi, the coding agent Orbi uses, is not already installed, you also need Node ≥ 22.19 and npm.

01 · INSTALL
Install Orbi
The script installs uv if needed, uses it to install Orbi with Python 3.14, and installs Pi if missing. It checks out Orbi to ~/.orbi/src by default, then starts orbi setup. It does not install git, gh or your system’s scheduler.
02 · MODEL
Choose a model supported by Pi
Orbi uses Pi’s model connections. Use a Pi-supported provider with an API key or a supported account login; inside an interactive Pi session, /login lets you choose an account provider. Custom endpoints and local models use Pi’s supported protocols, which include more than OpenAI-compatible APIs. The Pi provider guide lists the options. If a connection needs a Pi extension, enable that extension in Orbi’s configuration.
03 · REPOSITORY
Connect the repository you want to work on
Follow setup to select the GitHub repository, save orbi.toml and configure scheduled operation. gh auth status checks GitHub CLI authentication; it does not check SSH access. Use the full getting-started guide for these steps.
04 · FIRST ISSUE
Deliver your first Issue
Create an Issue in the connected repository, describe the result and acceptance criteria, then add ai-ready. Orbi implements it and opens a PR; review and merge follow in later scheduled runs. Follow progress in the Issue and PR. Missing information, failed checks or required approval may need your attention before it can merge.

FAQ

Before you install it

01Where does my data live?

Configuration, task working directories and logs live on the machine running Orbi. Credential storage depends on the connection: API keys can use the installation’s .orbi/env; Pi account logins use ~/.pi/agent/auth.json by default. GitHub holds Issues, labels, PRs and releases. A hosted model also receives the code context sent for a task.

02Can I use a local model?

Yes, if Pi supports the model’s connection protocol. The Orbi provider guide includes a local Qwen template.

03Is this Claude Code's self-hosted environments?

No. Anthropic’s Claude Code self-hosted environments run Claude Code cloud sessions on your infrastructure while Anthropic hosts the control plane. Orbi uses Pi to deliver GitHub Issues through implementation, independent review and merge, subject to repository checks and required approvals.

04How do I pause new tasks?

On Linux, systemctl --user disable --now orbi@1.timer pauses future starts for the default instance; an already running task continues. Use your actual timer names, and pause every enabled instance. macOS uses launchd instead of systemctl. See Operations for the Linux timer and running-service checks.

05What does one Issue get?

A feature or fix Issue leads to a PR with implementation evidence and independent review. Orbi merges the reviewed commit once checks and required approvals pass. Request a release in a separate Issue: add ai-ready and ai-release, and state the target version. Orbi follows the repository’s release rules and creates a tag and GitHub Release after related tasks and release checks are complete. See Issue to release for the workflow.

06How does it compare with Devin?

See Orbi vs Devin for a comparison of setup, pricing and delivery workflow.

07What do Pi and Orbi each do?

Pi runs the coding and review sessions. Orbi selects Issues, starts those sessions separately, follows task progress and handles merge and release decisions. Read How Orbi is built on the Pi coding agent for the division of work.

START WITH ONE ISSUE

Hand over your next Issue

Sign in to Orbi Cloud with GitHub, authorize the App and connect a repository to submit your first task. Orbi maintains the runtime while you follow implementation, independent review and merge in GitHub. Request a release separately when the changes are ready to ship.