GitHub-native software production
No new workspace. Orbi runs the delivery line on the Issues already in your repository. GitHub stays the source of truth.
Register with GitHub and continue onboarding in Cloud.
Prefer to self-host? Run Orbi yourself ↗
Inspect how Orbi builds Orbi →PRs merged by the machine · releases shipped · unattended since
orbi-build/orbiOrbi itself
orbi-build/orbi-websiteThis website
orbi-build/orbi-cloudPrivateCloud control plane
Teams already running Orbi
THE SHIFT
As agents work for hours and run in parallel, the bottleneck moves. It is no longer typing code. It is deciding what should change, ordering dependent work, proving the result, controlling authority, and carrying an exact version into production.
Stronger agents make the production system more important, not less.
SHIPPING TODAY
An Issue is the executable unit, but Orbi does not manage Issues in isolation. Epics and milestones hold scope. Dependencies order the work. Review and repair converge each PR. A Release Issue freezes the result into a version.
Group outcomes without letting the Runner mistake coordination for execution.
Use GitHub's real dependency graph. A task enters production only when its blockers close.
A second session may patch the branch. Only its reviewed head is allowed to land.
Validate scope and the repository, then publish an immutable version—without an Agent improvising the release.
Process killed. Test failed. Commit not pushed. The same run, branch, worktree, and PR remain the scene for the next tick.
PRODUCTION EVIDENCE
Issue #48 entered the real Orbi repository. An isolated run implemented it, a separate session reviewed the exact diff, PR #193 merged, and the work shipped in Release v0.2.0 with every Issue in scope verified item by item. The record is still inspectable because the record is GitHub.
The reviewer changes code, reruns the suite, and judges the repaired head.
A fresh base, a mergeable PR, and the exact commit that received the verdict.
Restarts reconstruct the delivery from the Issue, PR, labels, journal, and run ID.
Run this on your own repository
curl -fsSL https://aiready.sh | shNeeds Python 3.14, git, gh, systemd, and a model endpoint. orbi doctor names whatever is missing.
git clone https://github.com/orbi-build/orbi.git && cd orbi
uv tool install --force --reinstall --editable --python /usr/bin/python3 .
cp .orbi.example.toml orbi.toml
orbi setup --config orbi.tomlPrefer to self-host? Run Orbi on your own machine — the open source (AGPL-3.0) core is free forever. Setup guide ↗
YOUR FACTORY FLOOR
Orbi runs where the repository and model access already live. GitHub carries durable coordination and evidence; your machine carries the execution.
RUN IT YOUR WAY
The self-hosted, open source (AGPL-3.0) core ships today. Managed Cloud is Orbi's commercial managed service on the same GitHub-native ledger — with a founding partner offer.
THE DIRECTION · NOT A SHIPPING CLAIM
Today, explicit GitHub tasks and Release Issues enter the line. The next horizon is a closed production loop: software supplies evidence from its own operation, Orbi turns that evidence into bounded work, Agents execute, and the result is measured after release.
Incidents, telemetry, user feedback, security findings, dependency drift.
Ground work in evidence; order it by goals, dependencies, risk, and cost.
One Agent for sequential work, coordinated specialists where work can truly split.
Tests, evaluations, simulations, security rules, and explicit approval boundaries.
Immutable releases, progressive delivery, automatic rollback, complete provenance.
What worked becomes evidence. What failed becomes a reproducible task—not folklore.
Models will change. Agents will improve. The durable product is the layer that keeps intent, state, authority, verification, economics, and production evidence coherent across all of them.
Read the sourced competitor comparisons →
Orbi should not become another chat window or another board. It should become the control system above the Agents.
BEFORE YOU INSTALL
Git, a POSIX shell, and API access to a model you already pay for. Orbi runs on Linux and macOS, on your own machine or your own server. No GPU is required — the model does the thinking, and it can be a hosted API.
Self-hosted: your repository, keys, and test suite stay on your machine; nothing is uploaded to us. Cloud: the runner clones the repository you authorize onto machines we operate for the duration of a delivery — see “Can you see my code?” on the Cloud page.
Bring your own. Orbi drives the agent through your credentials, using any OpenAI-compatible API or a locally hosted model, and you can change your mind later without changing your workflow.
Model spend goes to your provider at your rate. Orbi charges nothing on top: the self-hosted core has no platform fee.
Those are editors — they make you faster while you are typing. Orbi is what happens when nobody is typing: it takes an Issue from your backlog, works in an isolated worktree, and takes it through review and merge to a tagged release.
The difference that matters is what happens after the code is written. A second session reviews the exact diff, repairs it, and reruns the suite, and only the reviewed commit is allowed to merge. That review step is the product.
Another Orbi session, running independently of the one that wrote the code, and it can change the branch rather than only comment on it. It reruns the tests against its own repairs and judges the repaired head.
You stay the last gate. The work lands as a normal GitHub pull request, so your branch protection, required checks, and human approvals apply exactly as they do today.
The next tick returns to the same scene. Because the Issue, the pull request, the labels, and the run ID live in GitHub rather than in a process, a killed process or a failed test does not produce a second story about what happened.
Restarts reconstruct state from that record — the same run, branch, worktree, and PR.
Open source under AGPL-3.0, or the Sustainable Use License if your policy rules out AGPL. Self-hosting is free under either.
Managed Cloud starts with GitHub: sign in, install the App, and subscribe. Paid annually, Solo is US$24/month (US$290/year, save 17%; monthly billing US$29) with 100M tokens and Pro is US$66/month (US$790/year, save 17%; monthly billing US$79) with 300M tokens. Founding partners get 50% off forever, limited to 6 places.
OPEN SOURCE (AGPL-3.0) CORE
Open source (AGPL-3.0). Self-hosted. Bring the model you already trust. Keep GitHub as the record, and let Orbi carry the work to a tagged release.
THIRD-PARTY DELIVERIES
Merged in other people's repositories
Orbi opened these pull requests in repositories it does not own. Every link opens the public GitHub record.
Tommy Xiao @xds2000
xiaods/k8e
M0：兼容规范和 rqlite 能力验证
xiaods/k8e
test(etcd): 设计并分阶段落地嵌入式 etcd 健壮性 E2E 方案
SHUKE-LABS/mat-site
Replace terminal logo and favicon with new four-agent My AI Team artwork
no human review
SHUKE-LABS/mat-site
Let mat-site use the same slogan as https://shukelabs.com/products/my-ai-team/ does
no human review
zzuu080603/Tianshu-harness
buildOaiRequest 每轮 5+ 趟全量扫描的性能战役（战略票）
no human review
5 deliveries in other people's repositories →