The same /optimus command works in the Browser Agent and in Unlok Code. Ask a question once, or toggle a mode so every message you type is a question until you turn it off. This page covers exactly how to use it in each place.
Both surfaces support the exact same two commands. Nothing to configure: if you're signed in to Unlok, they just work.
/optimus what's our deploy process?
/optimus
Sending /optimus with nothing after it turns the mode on. A small banner appears above where you type so it's never ambiguous whether your next message is a question or a task instruction. Send /optimus again, or use the banner's Exit button, to turn it back off. Any other slash command your extension supports still works as itself while the mode is on; only plain text gets redirected to Optimus.
This is the part that matters most. If an agent is actively working (editing files in Unlok Code, browsing in the Browser Agent) and Optimus answers, the answer doesn't just print on screen for you to read. It gets fed straight into that task's own conversation, the same way a message you actually typed would be, so the agent's next action actually accounts for what you just learned.
With no task running, the answer just shows up as a normal message. Nothing else to do. Either way, you never have to stop what the agent is doing to ask something on the side.
One thing worth knowing: once an answer is fed into a running task, it becomes part of that task's real context on its next step. A short answer costs nothing worth noticing, but a very long one mid task does add to what's already running.
Open the side panel and type into the chat box like any other message. If a browsing task is currently running, you'll see the normal progress indicator; asking Optimus shows its own short loading row while it thinks, which clears the moment the answer (or an error) arrives.
/optimus does our checkout service use Stripe or Paystack?
/optimus what did we decide about rate limits last week? and what repo is that in? /optimus
A digest of your (or your team's) saved memory and connected repos gets built the first time you ask, and reused for about 20 minutes after that. The first question in a while takes a few real seconds; every question after it is close to instant.
Works the same way in the chat composer. While a digest is being built you'll see a small “Asking Optimus…” row with a spinner above the input, the same treatment used elsewhere in the extension for a request with no streaming progress of its own.
/optimus what did we decide about the auth migration?
/optimus
If a task is running when the answer comes back, it's injected as a real turn in that task's own conversation. It reads like a normal message in the transcript, not a special notification, and the agent continues from there with that context in hand.
/optimus isn't a built in CLI command yet; it's next on the list. Until then, the same digested memory is reachable directly over HTTP with your own Unlok API key:
curl https://unlok-backend-xpts.onrender.com/v1/me \ -H "Authorization: Bearer $UNLOK_API_KEY"
The response includes a user_id field. Save it; it doesn't change.
curl https://unlok-backend-xpts.onrender.com/v1/optimus/YOUR_USER_ID \
-H "Authorization: Bearer $UNLOK_API_KEY" \
-H "Content-Type: application/json" \
-d '{"question": "what repos do we have connected?"}'Find your Unlok API key on the dashboard's API keys page (or reuse the one unlok auth already connected, if you use the CLI too). The same rules apply as everywhere else: you can always ask about your own account, and your whole team's memory too if your role already grants that on the dashboard.
| Command | What it does |
|---|---|
/optimus <question> | Ask one question, once. |
/optimus | Toggle Optimus mode on or off. While on, plain messages are questions. |
| Mode banner's Exit button | Turn Optimus mode off without retyping the command. |
Optimus answers from a digest of your saved conversation memory, plus your connected GitHub repos. Basic repo info (name, language, README) applies to every connected repo; for repos with real code access turned on, Optimus also gets a short overview of what the code actually does. If you're on a team and your role allows it, the same digest pools everyone's memory instead of just your own.
The digest is what makes repeat questions fast: it's built once, then reused for about 20 minutes or until new memory is saved, whichever comes first. Asking several things in a row after the first one costs almost nothing.
On a team workspace, Optimus also follows the workspace's rules and policies marked for it. They are read once, when the digest is built, and shape every answer until it refreshes, whichever product asked. Each answer reports the rules version it ran under. See workspace rules and policies for how rules are written and where else they apply.
The operational layer between your team and every AI model it touches.