Open Codelit and describe what should be true when the work is finished. Codelit chooses what to create, asks one focused question only when the destination is genuinely ambiguous, and keeps every result in the same Thread.

Or choose Check a website, Research something, or Plan a project under the empty composer. The next question appears in your Thread: reply with the website address, topic, or project brief. You do not need to choose an app or model first. These suggestions disappear while you type and return when you clear the composer.

1. 1Describe the result in one message, including a URL, destination, or schedule only when the work needs one.
2. 2Review the result in the conversation. Choose Open for full visual control in the same Thread, or ask for a correction without starting over.
3. 3Keep working in the same Thread. Agent Teams expose Flow, safe proof, Live run, Apps, and advanced settings in their native workspace; Back to Thread restores your exact place in the conversation.

Ask Codelit to inspect one website and summarize its findings. Browser handles the website task directly; no Agent Team is required. After a result is available, ask to send these findings to Slack, create a GitHub issue, or save a report to Files. Complete any missing access step, then review the exact destination and outgoing content before approving an external action.

Open the Agent Team in the Thread and choose Run for its safe proof first. It checks committed example inputs, evidence, and approvals without calling a model or external App. Use Run options only when you intentionally want Live run and its account, connection, scope, approval, and spend checks.

• Run progress and the receipt appear in chronological order in the Thread.
• Approvals stay inline instead of freezing the design behind a modal.
• Move to live data only after the safe proof is useful.

Ask Codelit carries your request from proof to live work without leaving the Thread. Codelit reuses a healthy App and the Team's previous resource when it remains valid, automatically selects a sole safe resource, and asks one compact question only when several choices are plausible. Browser evidence, terminal output, approvals, failure recovery, and the final receipt stay attached to the owning Agent Team.

• Connector writes and browser actions stay simulated during the personalized proof.
• Trigger Hub chooses the safest compatible starting mode and keeps schedules, limits, health, and receipts together.