Chat

Chat uses the same wallet credit as your API keys. Connect a wallet with an Agent ID, sign in, and make sure it has enough available credit for a request.

Start a conversation

Choose New chat in the workspace sidebar. Your first message creates a conversation and supplies its initial title. The suggested prompts fill the composer so you can edit them before sending.

Select a model in the bottom composer. The model menu shows its input and output prices per million tokens and maximum output length. Type a message and press Enter to send. Shift + Enter adds a new line. Messages accept up to 8,000 characters.

Replies appear when the request completes. Use the copy action on a reply to copy its text. The scroll-to-latest control returns you to new messages when you have scrolled up.

Choose a thinking level

Auto uses the provider's default behavior. Low, Medium, and High are selectable only if the model declares support for those effort levels. Changing the model resets thinking to Auto. If capabilities cannot be checked, use Auto or retry the capability lookup.

More reasoning can take longer and consume more output tokens. Those tokens count toward the model's output limit and billable usage even though the interface only displays the final answer. Thinking does not expose the model's private reasoning.

Manage local history

The sidebar lists your conversations. Open a conversation to restore its messages, draft, model, and thinking level. Rename it from its menu or delete it when it is no longer needed. Deletion removes the local conversation and cannot be undone.

History is saved in localStorage for the signed-in wallet in this browser. It does not sync to another device or browser. Signing out or changing wallets hides the previous account's conversations; signing back in with the same wallet restores them. Signing out does not erase browser storage.

Clearing site data removes local history. If browser storage is blocked, unreadable, or full, an inline message explains that changes may not survive a reload. Save important answers elsewhere before clearing storage.

Context and credit

Each request sends up to the last 20 eligible messages along with an AgentPass system prompt. Earlier messages can remain visible locally without being included in the next model request. Prompts and recent context go to OpenRouter and the selected provider. See Legal and privacy.

Open Chat information above the conversation to check your balance and recent usage or navigate to Top up. Credit is reserved before generation and settled afterward. Reasoning tokens are billable output. If a successful response lacks valid output usage, Chat uses the reserved maximum output for billing, capped at the reservation. See AI credit.

Retry a failed message

A failed or interrupted message can be retried. Retry creates a new request and uses the selected model and thinking settings. An already accepted request can finish and use credit even after you close the page or delete its local conversation.

For an identity error, check the wallet and registry. For insufficient credit, top up or choose a model whose reservation fits your balance. A provider failure releases the reservation. If the service reports a rate limit, wait before sending again; the wallet limit is shared with API keys.