OpenClaw Control UI: a practical workspace for parallel agent work
OpenClaw Control UI is the browser workspace for running and supervising an OpenClaw Gateway. It is useful when one agent task is no longer enough: you need to inspect a background task, compare two sessions, check context pressure or usage, and still keep approval controls close. OpenClaw v2026.7.1 makes that workflow much more deliberate with resizable session panes, live Tasks, clearer usage data, and mobile pairing from the UI.
The important distinction is that this is an operator surface, not a public chat widget. The Gateway serves it locally by default, authenticates it during the WebSocket handshake, and may require device pairing for a new browser. Treat it like a console with real authority.
Contents
- What the OpenClaw Control UI is for
- How parallel sessions change the operator workflow
- Use Tasks and usage data to decide when to intervene
- Keep the Control UI private and paired
- A practical operating routine
- FAQ
What the OpenClaw Control UI is for
The OpenClaw Control UI is a single-page app served by the Gateway. On a local Gateway, the documented entry point is http://127.0.0.1:18789/; openclaw dashboard can reopen a clean dashboard URL and provide an SSH hint on a headless host.
Its job is operational visibility and control. The UI can put conversations, settings, device pairing, approvals, Gateway status, workspace files, scheduled jobs, and background work in one place. That matters once the agent is doing work that outlives a single chat turn. A terminal remains useful, but it is a poor place to compare active sessions, see an interrupted run beside its replacement, or follow a task while another session is waiting for approval.
OpenClaw’s own overview of how the Gateway works is helpful context here: the UI is connected to the Gateway, not a separate assistant with its own hidden state. The benefit is that an operator can see the same running system that channels and agents use.
| Need | Useful Control UI capability |
|---|---|
| Resume or compare work | Find, pin, group, rename, fork, archive, and mark sessions read |
| Watch background work | Open the live Tasks view, inspect its session, refresh details, or cancel authorized work |
| Understand spend and context | Review provider, model, agent, or channel usage alongside tokens, cache use, context pressure, and estimated cost |
| Connect a phone safely | Create a short-lived mobile setup QR or code from an authorized paired administrator session |
| Change runtime settings | Use the config form backed by the live config schema, with Raw JSON available when needed |
How parallel sessions change the operator workflow
The standout v2026.7.1 addition is not a new model picker. It is a better way to keep more than one piece of work visible without losing your place.
The Control UI can show multiple resizable sessions side by side or stacked, preserve the layout after reload, and let an operator drag a session into a new split. That gives parallel work a visible shape. One pane can hold a research or planning session; another can hold the execution session; a third can expose a task that is waiting on a tool, a channel delivery, or an approval.
This is different from opening more browser tabs. Tabs hide the decision context. A pane layout makes it easier to compare the current instruction, tool output, and agent status before you redirect a run. It also reduces a common failure mode in long-running agent work: the operator remembers that something was active but can no longer tell which conversation owns it.
If your system uses multiple agents, pair this with the OpenClaw multi-agent setup guide. The UI does not replace explicit task boundaries or workspace ownership. It gives the person supervising those boundaries a clearer place to inspect them.
Use Tasks and usage data to decide when to intervene
A dashboard only helps when it changes a decision. v2026.7.1 adds a live Tasks page for active and recent background work. Authorized operators can open the task’s session, refresh its details, and cancel queued or running work. That makes it easier to distinguish a slow task from a task that is genuinely stuck.
The release also expands the Usage page. It can compare estimated cost and daily values, break usage down by provider, model, agent, or channel, and show provider-reported information such as plans, quotas, balances, budgets, or reset times when those integrations are configured. Chat-level context details can expose the active model, token usage, cache use, context pressure, and estimated cost.
Use those signals together rather than treating cost as a scorecard:
- Start with the task state. Is it progressing, waiting for a tool, or waiting for approval?
- Check the owning session. Does the conversation show a clear next action or an error that needs a human decision?
- Check context pressure and usage only after you know the task’s state. A costly task may be valid; a quiet stalled task may be the real problem.
- Intervene with a narrow instruction, a cancellation, or a new session only when the UI gives you evidence for it.
That is close to the durable-control-surface principle described in AI agent UI state: the operator needs a reliable view of the session and active work, not merely a transcript after the fact.
Keep the Control UI private and paired
The Control UI can create sessions, manage configuration, and expose approvals. Do not publish it as if it were an ordinary website.
For a local Gateway, use the loopback URL. For remote access, the OpenClaw documentation recommends a protected route such as Tailscale Serve, a trusted identity-aware proxy, or an SSH tunnel. The dashboard documentation also explains that tokens are kept in the browser’s current session storage rather than local storage, and passwords are not persisted across reloads.
New browsers and devices can require a one-time pairing approval. A request to broaden an already paired browser from read access to write or admin access is an approval upgrade, not a silent reconnect. Review the requested role and scopes before approving it. The mobile setup code is short-lived and should be handled like a password while it remains valid.
These are not cosmetic safeguards. They keep a convenient operator workspace from becoming an exposed control plane. For the larger self-hosting tradeoff, see Is OpenClaw safe?.
A practical operating routine
The OpenClaw Control UI is most useful when it supports a repeatable habit:
- Open the dashboard locally with
openclaw dashboardor the loopback URL. - Keep the active execution session visible, then add a second pane only when comparison will change a decision.
- Use Tasks to locate background work before assuming a run failed.
- Check the session’s context and usage details when a task needs a routing, scope, or budget decision.
- Review pairing requests and approval scopes before granting broader access.
- Close or archive work that is done so the session list stays useful.
The v2026.7.1 release does not remove the need for clear goals, narrow tool permissions, or safe remote access. It makes those controls easier to use while the work is still happening.
FAQ
What is the OpenClaw Control UI?
It is the browser-based operator interface served by the OpenClaw Gateway. It connects to the Gateway over WebSocket and provides access to sessions, tasks, settings, approvals, pairing, status, and other administrative controls.
How do I open the OpenClaw Control UI?
For a local Gateway, open http://127.0.0.1:18789/ or run openclaw dashboard. If the Gateway is remote, use an authenticated private route such as Tailscale Serve or an SSH tunnel instead of exposing the UI publicly.
Can the Control UI show more than one agent session?
Yes. In v2026.7.1, sessions can appear in resizable side-by-side or stacked panes, and the layout can persist after a reload. This is useful for comparing active work without shuffling browser tabs.
Is the Control UI safe to expose to the internet?
No. It is an admin surface. Use Gateway authentication, device pairing, and a private remote-access path. Review any request to upgrade a paired browser’s scope before approving it.