Chats

Chats are how you talk to Stoa. Most of your work happens here: asking questions, drafting documents, reviewing reports, getting cited answers from your organization's knowledge.

Starting a chat

From the main navigation, click New chat. You'll see an input box, the agent picker, and a mode toggle.

• Agent picker. Choose the agent for the conversation. Each agent has its own knowledge bases, tools, and personality. If you only need general Q&A, the default agent is fine.
• Mode selector. Four modes: Auto decides per message whether to reason before answering, which is right for daily questions and quick tasks. Instant always answers straight away, with no reasoning step. Thinking always reasons before answering. Research runs the longer, structured pipeline described in Deep research, and is chosen fresh each time rather than staying on as your default.
• Input box. Type your question in plain language. Stoa understands natural questions, you don't need keywords or special syntax.

Press enter and Stoa starts working.

What happens during a chat

In Auto mode, Stoa:

1. Reads your question.
2. Decides which knowledge bases and tools are relevant.
3. Searches and calls them as needed.
4. Composes an answer with citations.
5. Streams the response to you in real time.

You can see the reasoning as it happens: which documents were retrieved, which tools were called, and how the answer is being assembled.

In Research mode, Stoa runs a longer, multi-step pipeline. See Deep research.

Citations

Every answer that uses your documents shows a Sources button underneath it. Open it to see exactly which passages the answer drew from, and where in each document they came from. See Citations for how to read and verify them.

If you don't see a Sources button on an answer that should have one, the answer wasn't grounded in a knowledge base. That's normal for general-knowledge questions, but it's a flag for compliance-style questions where you expect a documented source.

Following up

You can ask follow-up questions in the same chat and Stoa keeps the context. "Who else does this apply to?" "What was the date on that document?" "Can you summarize that in three bullet points?" all work without restating the original question.

Copying answers

Hover over an answer and click the copy icon to copy the rendered text to your clipboard. Useful when pasting into a report, an email, or a slide.

Feedback

Every answer has a thumbs-up and thumbs-down button. Use them. Feedback feeds the quality loop your administrators monitor: thumbs-down responses get reviewed, prompts get refined, and answers get better over time.

If something is factually wrong, downvote it. If something is great, upvote it.

Chat history

Your past chats appear in the sidebar. You can reopen any of them, continue the conversation, or delete the ones you don't need anymore. Search the sidebar to find a specific chat by content.

Privacy

Chats are stored in your organization's Stoa deployment. They never leave your network. Your administrators control retention policies.

If your organization uses commercial model APIs (like OpenAI or Anthropic) for some agents, the prompt for those specific agents is sent to that provider for that one call only. Most deployments use locally-hosted open-source models for everything, in which case nothing leaves your infrastructure at all.

Where to next

• Knowledge bases for how documents become searchable.
• Agents for how to pick the right assistant for your task.
• Deep research for in-depth pipelines.
• Citations for how to read and verify them.
• Feedback for how your votes shape Stoa over time.