Chat

The dashboard Chat is a governed question-answering surface over your organization's memory. Every answer runs through the same policy gate as the recall API — it can only cite memories your role and level are cleared to see.

How it works

1. You ask a question.
2. The agent calls the recall tool (and get_fact for deterministic facts), scoped to your authenticated role — shown as policy:allow · N memories retrieved or a policy:deny badge above each answer.
3. The answer is rendered as markdown with inline [n] citations, followed by a Sources list.

Suggested prompts

The chips under the input on a fresh chat are not AI-generated — they are your most recent memory titles, filtered to the scopes and clearance level you can access (GET /api/memories/suggestions). They refresh each time you open the page, so they always reflect what's actually in your governed memory.

Citations

• Inline [1], [2] markers in an answer are clickable — selecting one expands its source.
• Each Source card can be expanded to preview the memory, with an Open in Memories link to the full memory detail page.
• A source card only appears for a memory the answer actually cited — retrieved candidates that weren't used as evidence are not shown.

History

Open the history panel (the clock icon) to browse Recent Chats. Selecting a conversation reloads its full message thread; New Chat starts a fresh one. Reloaded answers render with the same governed layout (policy badge + sources) as the live stream.