Chat
The conversational interface over your knowledge base and agents. Ask questions, invoke agents, run workflows, collaborate — all from one chat surface.
What chat is (and isn’t)
Chat in StellarBase is not a single general-purpose AI assistant. It’s a dispatcher — your message gets routed to the right agent or workflow based on context, and the agent’s grounded response is returned. When you say “summarise the Q2 beneficiary reports,” chat invokes your ESF+ Reviewer agent; when you say “find me the Müller paper on surface codes,” chat invokes the search path.
The result: you don’t have to know which agent does what. You just ask.
Threads & Bases
Every chat lives in a Base. The agents, data, and permissions of that Base bound what chat can do. Threads keep context across multiple messages; you can reference earlier messages (“refine the second paragraph of the draft above”).
Chats can be shared with colleagues — either read-only or as joint threads where multiple people and agents collaborate. Every message is part of the audit log.
Mentioning agents
By default chat routes intelligently. If you want to force a specific specialist, @mention the agent by name. Examples:
- “@MSA-Reviewer check the Lipsia contract for CoC issues” — direct invocation
- “@Research-Agent find recent papers on surface codes, then @Writing-Assistant draft a summary” — chain two agents
- “Compare contract 0042 with contract 0088” — no mention, chat picks the contract agent
Citations & sources
Every response from chat includes citations to the underlying sources. Click a citation to jump to the exact passage in the document. If you ask chat to produce a claim, it attaches the evidence automatically — if no evidence exists, chat will say so rather than invent.
Chat is configured by default to refuse to answer when no grounded evidence exists in the knowledge base. This can be overridden per Base, but it’s the safer default for regulated work.
Tools from chat
Any tool registered with your agents is also callable from chat — directly or through an agent. Common patterns:
- “Translate this clause to English” — chat calls the translation tool
- “Export all HIGH-risk contracts to CSV” — chat invokes the export workflow
- “Run policy-comparison against the new Generali quote” — chat triggers the named workflow
- “Book a meeting with the legal team for Thursday” — chat calls the calendar connector
Files in chat
Drop files directly into chat — PDFs, images, spreadsheets. They’re ingested into the Base, and chat immediately has access. Useful for one-shot questions about a document that doesn’t need a permanent place in the corpus.
Conversation memory
Chat maintains context within a thread. Across threads, persistent facts (your name, your role, your preferences) are stored in a per-user profile that you control. You can inspect and edit what chat remembers at any time.
Voice
Chat supports voice input/output where the deployment includes the optional speech-to-text and text-to-speech models. Useful for hands-free workflows in clinical or field settings. See Specialized Models.
Privacy
Chat respects Base permissions end-to-end. It won’t surface documents a user can’t access, even if they’re semantically relevant. When chat calls an external LLM, the prompt is anonymized via StellarGate first — no PII leaves your perimeter.
Related
- Agents — what chat dispatches to
- Workflows — longer-running pipelines triggered from chat
- Search — the retrieval layer under chat’s hood
- Collaboration — shared chats and review flows