Appearance

The chat interface is how you interact with agents in Frona. You type messages, agents respond in real-time, and you can see every tool they use along the way.

Starting a conversation ​

  1. Open the chat interface
  2. Select an agent from the sidebar. The Assistant is selected by default
  3. Type your message and press Enter
  4. The agent's response streams in real-time, token by token

Each conversation is linked to one agent. To talk to a different agent, start a new conversation.

You can also start conversations from a connected channel. Sending a message to a Telegram bot or texting a paired SMS number creates a conversation in the channel's space. See Channels.

What you see during a conversation ​

Real-time responses ​

Agent responses appear as they're generated. You see tokens arrive in real-time. This means you can start reading before the agent finishes responding.

Tool activity ​

When an agent uses a tool (like searching the web, running code, or reading a file), you see it happen inline:

  1. The tool name and what it's doing appear in the chat
  2. The tool runs and its results are displayed
  3. The agent continues its response using the results

Agents can chain multiple tools in a single response. For a complex request, you might see the agent search the web, read several pages, and then synthesize the results, all visible step by step.

Usage and cost ​

Once an agent has made its first model call, a pill in the chat header tracks the conversation's running token count, cost, and how full the agent's context window is — updating live as you go. See Tracking Usage & Costs for the pill and the full usage dashboard.

Interactive moments ​

Sometimes an agent needs your input before it can continue. The chat pauses until you respond, then resumes from exactly the same point.

  • Questions. The agent presents you with options to choose from. For example: "Which format do you want the report in? PDF / CSV / Markdown." Click your choice and the agent proceeds.
  • Approvals. Before deploying an app or doing anything else that needs a green light, the agent shows the proposal and asks "Yes / No". One click and the agent either runs the action or backs off.
  • Credential picks. When the agent needs a secret, it pauses with a vault picker so you can choose which entry to grant. See Managing Secrets & API Keys.
  • Browser takeover. If the agent encounters something it can't handle (like a CAPTCHA or two-factor authentication), it pauses and gives you a link to take over the browser session. Handle the interaction, then let the agent continue.
  • Task completions. When a delegated task finishes, the result appears in your conversation automatically.

The same prompts appear when you're chatting from a channel: Telegram, Discord, Slack, and WhatsApp Cloud show real buttons; SMS, Signal, and WhatsApp Personal append a "Reply YES or NO" hint and parse your reply.

Slash commands ​

Type / or @ at the start of the composer to invoke a command:

  • /new — start a fresh chat with the same agent.
  • /clear — clear the current chat history.
  • /<skill> — invoke an installed skill directly (e.g. /weather Lisbon). The skill runs even if the agent wouldn't have picked it on its own.
  • @<agent> — hand the current message to another agent for one turn (e.g. @developer fix this error). The reply is attributed to that agent; the next message reverts to the chat's default.

The composer shows a popover with everything available, including each item's argument-hint. Channels parse the same prefixes, but they don't get autocomplete — you have to know the name.

Context and memory ​

Each conversation draws from several sources of context:

  • Message history. Everything said so far in this conversation
  • Your memory. Facts the platform remembers about you (preferences, background, etc.)
  • Agent memory. Facts specific to this particular agent
  • Space context. If the conversation is in a space, context from previous conversations in that space

When a conversation grows long, the platform automatically summarizes older messages to keep things running smoothly. Recent messages stay in full. Only the oldest parts get compressed.

Attachments ​

You can upload files directly in the chat. Drag and drop a file or use the attachment button. The agent can then read and work with the uploaded file.

Agents can also produce files for you. These appear as downloadable attachments in the chat.

Managing conversations ​

  • Sidebar. Your conversations are listed in the sidebar, most recent first. Click any conversation to switch to it.
  • Archive. Archive conversations you're done with to keep the sidebar clean. Archived conversations can be restored later.
  • Titles. Conversations get auto-generated titles based on the first few messages. You can rename them for easier navigation.
  • History. Long conversations load the most recent messages first; scroll to the top to fetch older messages in pages. Day separators mark date boundaries, and longer gaps between messages get their own marker. Hover any message to see its exact timestamp.

Tips for better results ​

  • Be specific. "Research competitor pricing for project management tools and format as a comparison table" works better than "look into competitors"
  • Break down complex requests. List the parts clearly, or let the agent delegate sub-tasks to specialist agents
  • Use the right agent. The Assistant handles general tasks, but the Researcher is better for web research and the Developer for coding
  • Provide context upfront. Tell the agent what it needs to know rather than expecting it to guess
  • Use spaces for ongoing projects. Start conversations in a Space so agents carry context from previous conversations

Next steps ​

Frona AI