Skip to main content
Every programmable agent run is a session, whether it was started from a chat in the app, a Bolt schedule, Slack, or the API. The DinoAI Agent app collects them all in one place, so you can follow a run live, read back what an agent did, and stop one that has gone off track.

​
Starting a new chat

New chat at the top of the Agent app’s left panel opens a page with just the composer. Pick an agent from the dropdown above the input, or leave it on Default agent to start a session with no agent definition, then type your prompt and send. The page swaps to the live transcript as soon as the session starts. Hovering an agent card on the Agents page and clicking Chat lands on the same page with that agent preselected.

​
The sessions panel

The left panel of the Agent app lists your workspace’s recent sessions under Recents, newest first. Each row shows:
  • Source icon: where the session was started (chat, Bolt, Slack, or API).
  • Title: the opening message or a generated label. Hover a row for the source name and the full title.
  • Live indicator: a dot on sessions that are currently running, so you can tell an active run from history at a glance.
The filter control on the panel header narrows the list by Status, Type (chat, Bolt, Slack, API), or Agent. Sessions are deep-linkable: every session has its own URL you can share with a teammate.

​
Searching all sessions

The View all arrow on the panel header opens the full session list as a page. It adds what the panel can’t fit:
  • a search box that matches both session titles and session ids;
  • faceted filters for Status, Type, and Agent;
  • a sortable Started column and pagination for browsing history.
Use it to find an old run by id when debugging, or to filter down to one agent’s sessions across every source.

​
The session page

Opening a session shows a chat-style transcript with a compact header:
  • Status and session ID next to the title.
  • A pill strip with the source, start time, and credits consumed, plus runtime and burn rate while the session is still running.
  • A Stop button on running sessions. Stopping ends the run; the session stays in the panel as stopped rather than disappearing.
In the transcript, your prompts render as chat bubbles and the agent’s tool activity is collapsed into expandable “Ran N tools” lines, so the thread stays readable while the full detail (tool inputs and terminal output) is one click away. The thread sticks to the latest message while a run is live.

​
Continuing the conversation

Every session has a reply box under the transcript, no matter where it started: chat, Bolt, Slack, Teams, or the API. Sending a message starts the agent again with everything it already knows, so you can refine the task, ask follow-up questions, or ask a Bolt run about what it just did. The agent works through one message at a time: a message sent while the agent is busy is queued and picked up when the current turn finishes. A few source-specific notes:
  • Slack and Microsoft Teams sessions stay in sync in both places: a reply you send from the app is mirrored into the original thread under your name, and thread replies show up in the app transcript.
  • Bolt sessions are one run per trigger: each schedule run starts a fresh session, which you can then reply to. See Triggering from Bolt.
  • API sessions can also be continued programmatically with sendDinoaiAgentMessage. See the API Reference.

Related topics

Agent SessionsStarting a Slack Agent sessionReply to any agent sessionEvery agent session in one panelStart chats faster and search every agent session