documentation
03 dashboard

Chat

The chat page is your direct line to every agent. Pick an agent from the dropdown, send a message, and watch tool calls stream in as the agent works. This is where most of the ad-hoc interaction with Exolvra happens.

What this page is for

Chat is the conversational surface of Exolvra. Use it when you want a direct, interactive session with one specific agent — asking questions, running one-off tasks, testing how an agent behaves before giving it bigger work via issues. It’s the closest thing to a “just talk to an LLM” experience the platform offers, but with all the Exolvra scaffolding underneath.

Layout

The chat page has three things:

  1. The agent dropdown — top of the page. Shows every active agent (specialists and chatbots) grouped by kind. Pick the one you want to talk to. Switching agents mid-session starts a fresh conversation with the new agent.
  2. The conversation view — the main column. User and assistant bubbles stream top to bottom. Inline tool-call cards appear when the agent uses a tool, showing the call arguments and the result in a collapsible panel.
  3. The input box — bottom of the page. Type, press Enter to send, Shift+Enter for a newline. Attach files by clicking the paperclip icon.

If the selected agent is a chatbot with conversation starters configured, clickable starter chips appear above the input box on the first turn. Click one to auto-send it.

Common tasks

Start a new conversation

  1. Land on /chat — you’ll either see your most recent session or a fresh empty state
  2. Click the New chat button in the header (or use the dropdown to pick a different agent — switching agents creates a new session)
  3. Type your message and send

Resume an existing session

Open /chat/{sessionId} directly, or click any session from the Sessions page. All prior turns are restored; the agent sees its past messages and can reference them.

Attach a file to your message

Click the paperclip icon next to the input box, pick a file from your disk, and send. The agent sees the file’s contents (text files) or a reference to it (binary files) and can read, edit, or act on it depending on its tool set.

Switch agents mid-conversation

Use the agent dropdown to pick a different one. This starts a fresh session with the new agent — the previous conversation is saved and visible on the Sessions page. The new agent does not inherit context from the old one.

Share a chat

Every session has a permalink at /chat/{sessionId}. Copy the URL from your browser’s address bar. Anyone with access to your instance can open it and read the full transcript.

Tool calls in the stream

When the agent uses a tool, an inline card appears in the conversation. Click it to expand and see:

  • The tool name and arguments the agent sent
  • The result the tool returned (truncated if large, expandable to full)
  • Timing and cost for that specific call

Tool calls stream in real time — you can watch the agent do research as it does it.

What if the agent doesn’t respond well

Two common fixes:

  1. The agent’s personality is the issue. Open /agents/{id}/edit and tighten the system prompt. Be specific about tools to use, format to follow, and what “done” looks like.
  2. The agent’s tool set is missing something. If the agent can’t do what you’re asking, check its tool grants on the edit page. A research agent without web_search can only guess; granting the tool fixes it.

See Your first agent for the full agent-shaping guide.

Where to go next

  • Sessions — browse and manage past conversations
  • Agents — the agent roster and creator
  • Bots — the chatbot-specific gallery
  • Agents & chatbots — the conceptual difference