Agents fetch the requirements
Agents pull the task and the approved requirements themselves, and file the follow-up work they find back into Orneos.
For teams shipping code with AI agents
Review agent-written PRs against what you agreed.
Orneos keeps the agreement, your agents’ work and the pull request in one place. Agents start from the approved requirements, each captured run leaves a record, and your review starts from what was decided.
- MCP for agents
- Live agent activity
- GitHub PR review
The product
The app, in use.
A screen recording of the Orneos web app. One task is found, agreed and picked up, the board keeps working with the network cut, and the agent’s run is read back on the task. Then the same work on the timeline, in the team’s docs and in the numbers.
What changes
Most of the time now goes into checking what the agent did.
Writing code stopped being the bottleneck. Telling agents what matters, knowing what they really did and deciding whether the result is right is where the time goes now. Orneos is built for that part.
A record of what actually ran
Each captured run leaves a record on its task: which checks ran, on which commit, with which result.
Reviews open beside the agreement
The PR opens beside the approved requirements, pinned to the exact commit, with the gaps in plain sight.
Decisions tied to a commit
Every decision stays tied to the revision it covers, so the next person, or agent, starts from what was agreed.
How it works
Four steps: agree, deliver, watch, review.
-
Agree
Describe the task in your own words. Orneos AI drafts the goal, the acceptance criteria and what is out of scope, or you write them yourself. A person approves that version, and any later change needs approving again.
-
Deliver
Your agents connect to Orneos over MCP and start with the task, the agreement and the linked docs, each with its source. With your permission, they also file the follow-up tasks they find.
-
Watch
Runs are captured and linked to the task as they happen. The board shows where agents are working, and each run keeps what was observed apart from what the agent claims.
-
Review
Open the PR beside the agreement, pinned to the exact commit. Orneos AI sets out the evidence for each requirement and flags what is missing. You decide, and the decision stays traceable.
PR review
Each requirement, with its evidence or its gap.
Orneos connects to your GitHub repositories and opens the change pinned to the commit you are reviewing. Each agreed requirement shows its evidence, or the lack of it. Findings sit next to the code they cite. Looking at a file never counts as approving it, and your decision covers exactly the revision you inspected.
- PinnedEvery file and finding is tied to one commit; browsing another commit never swaps the one you decide on.
- AI-assistedOrneos AI sets out what supports each requirement and where the evidence stops, with its sources, so you start from the gaps.
- Visible gapsMissing checks and unresolved questions stay visible instead of disappearing into a green badge.
- Your callOrneos never edits code or merges. A person approves or sends back, and that decision is kept.
The workspace
Tasks, docs and reviews in one local-first workspace.
The tasks your agents read and the reviews you make sit next to your board and your docs, in the same workspace.
- Boards and backlogIssues, projects, timeline, labels, saved views and search.
- VaultNested pages with live co-editing and version history, linked from any task.
- Context bundlesTask, Intent, related work, comments and docs, assembled automatically.
- Local-firstEdits land instantly and are saved on your device before they sync; supported edits keep working offline.
Early access
Start with one team and one repository.
We’re onboarding teams one at a time. Bring the agents you already use, and we’ll set it up with you.
Orneos is built by Filip Ilievski, its founder. Early-access requests go to the founder directly.
Orneos is in early access, and capabilities reach teams in stages. See what’s live today
Questions
The short answers.
Which agents does it work with?
Claude Code and other agents that support MCP, the standard way agents connect to tools. You connect them with a personal access token. Agents can read tasks and agreements, and with your explicit permission they can also create and update tasks. Run capture starts with Claude Code.
Does it replace GitHub review?
No. The pull request stays in GitHub, and Orneos never edits code or merges. Orneos shows the change beside what was agreed and the evidence for it, and keeps your decision tied to the revision you inspected.
Does AI make the decision?
No. Orneos AI helps you draft the agreement and sets out the evidence for each requirement during review, using Anthropic’s models. It points out gaps; it never approves. A person approves or sends back, every time.
What is live today?
Orneos is in early access, and capabilities reach teams in stages. The feature catalogue lists exactly what is live. During onboarding we tell you what is switched on for your team, and where something is not yet, we work through it with you and say which parts were manual.
Do we have to leave our tracker?
Agents read and write tasks that live in Orneos, so teams usually start by running one team’s work there. Connections to other trackers are planned. See Integrations.
What does capture send from our machines?
Nothing, until a company admin allows sharing; sharing the agent’s final reply is a separate choice. Check results record the check you configured, its exit status and the commit state, never its arguments or command text. See Privacy and data.