Single-agent chat

When to use chat (vs synthesis)

Chat is the right tool when the answer should come from one perspective. If you want the CFO’s take on whether to raise prices, you ask the CFO. If you need the CFO + CMO + CRO + CEO to debate the same question, that’s a synthesis session.

Starting a conversation

Open Chat → New conversation. Pick:

• Agent role — the CxO seat. Defaults to the most recently used.
• Product — which topic-file library to ground in.
• Audience override (optional) — narrow the topic files the agent reads, e.g. only security + data-and-ai for a privacy review.

Each tenant message streams back via Server-Sent Events. The agent cites the topic files it read and any memos it referenced.

File citations

Citations are not decoration. Click one to open the cited topic file in the library side-panel. If the citation surprises you — that file says something different from what you expected — fix the file. The next conversation will read the correction.

Continuing a conversation

Conversations are durable. Open Chat in the sidebar to see them all. Re-open one and the model picks up the same role + product context. Anthropic prompt-cache reads make the second turn fast.

Quotas and limits

Per-tenant daily and monthly token caps apply. At 80% utilisation the workspace shows a warning banner; at 100% chat flips to read-only until the window resets. Cancellation of an in-flight message is best-effort — partial token usage still counts.

Saving a decision

When the conversation reaches a decision, click Save as memo. The memo captures: the question, the agent’s recommendation, the cited files, the model version, and any caveats. Memos are append-only — you can never edit a saved version, only publish a new one.

Audit trail

Every send, model call, memo save, file load, and quota event is logged to audit_logs append-only. tenant_admin users can review the full trail under Settings → Audit log.