Chat and conversations

Pazi conversations are working sessions, not one-off prompt boxes. You can guide an agent during execution, answer structured questions, inspect progress, and return to the same work later.

Main chat and work-specific chats

Each business has one visible Main Chat. Opportunities, delegated work, and scheduled work run independently in the background and return visible updates there. Slack and email keep their own channel history.

Use Main Chat for company coordination, opportunity plans, approvals, specialist questions, files, and follow-ups.

Send a request

Write the goal and useful context in the composer, attach relevant files, and send. The active agent starts work in that conversation and streams progress back to the dashboard.

Use Write a good request when the task needs clear constraints or a specific deliverable.

Send follow-up guidance

An ordinary follow-up sent during an active run is queued for that conversation. The agent finishes the current response, then processes queued messages in order.

This is useful for extra context that does not invalidate current work. It avoids interrupting an agent while it is saving files or completing a multi-step action.

Send now or stop

Use Send now on a queued message when the correction is urgent. Pazi interrupts the active run and starts that selected message immediately; other queued messages keep their order.

Use Stop when the current work should be canceled. Stopping clears the active run and queued messages for that conversation. Check any external side effects or files that may already have been created before restarting.

Answer structured questions

When an agent needs a choice or short answer, it can open a form above the composer. The form may contain single-choice, multiple-choice, or short open-text questions and shows your progress through them.

The regular chat input remains available while the form is open. Submitting the form sends a normal user message with the answer summary, so the decision remains visible in history.

Do not enter passwords, API keys, or tokens into a question form. Credentials use a separate secure flow.

Handle permission and login requests

The agent may ask you to enable web browsing, approve a domain/action, or sign in to a website through Live View. Read the requested action and destination before approving it.

Browser-permission approval can interrupt the blocked run so the agent can continue immediately. Saved domain rules can be reviewed later under account permissions. See Browser permissions.

Enter credentials securely

If an agent needs a secret, it should open a secure credential form instead of asking you to paste the value in chat. Saved values become available from the next run, not the process that was already running when you saved them.

See Integrations & credentials for storage, business scope, and deletion behavior.

Use voice

Voice calls and microphone input are currently available when the interface is set to English. Choose English in the language selector to start a call; text chat is available in every supported language.

Start a voice call from the dashboard when speaking is more natural than typing. The selected agent responds in its configured voice, and the transcript remains available with the conversation.

Voice is useful for discovery and redirection. For exact URLs, numbers, or long source material, follow up with text or a file.

Conversations in Slack and email

Slack threads and email threads are separate from the web conversation list. They use the acting agent's identity, skills, memory, workspace, and normal runtime but keep channel-specific history and routing.

Slack replies stay in Slack. Email from every agent appears in the selected business's shared Email Inbox, with outbound messages attributed to the acting agent. When a delegated agent or trusted email reply needs to return to the main web conversation, Pazi can surface it as an attributed chat card.

Recovering after a refresh

Web messages and the durable run state are stored by Pazi. Refreshing the dashboard should restore the conversation, active work state, questions, credential requests, generated images, file cards, and other recorded chat elements.

If a run has failed, read the visible error before resending. A fresh message can continue the same conversation, but do not assume an external action was rolled back automatically.