The workspace
The Onam Operations workspace is shaped like team chat and behaves like an operations console: plans, evidence, approvals and an audit trail around the conversation.
Status: Early access. Screens below are described from the design; the illustrations on the product page are mockups with sample data, not screenshots.
Three rules
| Rule | In the interface |
|---|---|
| Every claim is inspectable | Numbers and names in an agent’s reply open the evidence: skill, query, rows, scan and time |
| The plan is visible before it runs | A multi-step request shows its steps first, so you can stop it before it works |
| Actions look like changes, not messages | Proposed changes are distinct cards — diff first, with explicit decision buttons |
Navigation
| Group | Screens |
|---|---|
| Needs you | Home (what is waiting, running, failed) · Approvals · Inbox |
| Ask | Chat · Assets (the estate the agents see) · Agents (who does what, and may touch what) |
| Run | Tasks · Workflows · Reports |
| Govern | Knowledge (what you told the agents) · Activity (the audit trail and its chain) · Settings (autonomy, budget, kill switch) |
Chat
The primary working surface. Each agent answers in its own card, marked with its colour, so you always know who said what. You can target one agent (@Security), scope a question to accounts or environments, inspect evidence, and stop a running task at any time. A partially failed answer is never styled as a complete one: the unavailable domain is named.
Agents
The agent directory shows each agent’s purpose, level, status and skills. An agent’s detail page lists its skills, tools, permission tuples, knowledge sources and known limits — a trust surface, so you can see exactly what it may touch.
Tasks
Every unit of work, its state and what it is waiting on: running, waiting for approval, failed with a classified reason, cancelled or rolled back.
Approvals
The single queue where changes are decided — grouped by risk then age. Each card leads with the exact change, then blast radius, reason, rollback and impact. You can approve, reject (with a reason), modify within declared bounds, or simulate. See Governance and approvals.
Activity
The audit record: query by user, agent, task, resource, action or time, with a check that the hash chain is intact — no record edited or removed.
Reports
A conversation or task exported with its evidence preserved.
Settings and the kill switch
Enable or disable agents for your organisation, set the autonomy ceiling, budgets and notification routing. The kill switch stops all agent activity for your organisation; it asks you to type the organisation name and says plainly what will happen.