Chat

Chat is how you and your team interact with a cache in plain language. Ask questions, search across stored content, run workflows, and let the assistant write back to the cache on your behalf. Every conversation runs against the cache you have selected.

Starting a conversation

Open chat from the dashboard. Pick the cache you want to work in from the sidebar; chat runs against that cache until you change contexts. Ask in plain language; the assistant uses semantic search to pull relevant chunks before it responds.

Open Dashboard and select a cache to begin.

How context is resolved

When you send a message, the assistant builds its prompt in a fixed order:

Workspace System Rules — standing behavioral instructions.
Cache system instructions — context specific to the active cache.
Relevant chunks — the top matches from semantic search against the cache.
Blueprint instructions — present only when a blueprint run is active.
Your message and recent conversation turns.

The assistant draws from your cache and any connected tools. It does not pull from the open web unless a connected tool explicitly provides that capability.

Running blueprints from chat

Ask the assistant to run a blueprint mid-conversation. It resolves any required inputs, executes the workflow, and returns the result inline. For example: "Run meeting prep for my 2pm with Ana" or "Do inbox triage on what came in today".

If the blueprint requires an input that is not already supplied or resolvable from the cache, the assistant asks a follow-up question before running. See Blueprints for how inputs and triggers are configured.

Tool use

The assistant can take actions on your behalf using built-in tools:

Search the cache for relevant content.
Create, update, or move chunks when you ask for something to be captured or reorganized.
Read and write via connections to external services such as Gmail or Google Calendar when those are configured.
Call other blueprints or agents as part of a larger task.

Tool calls appear inline in the conversation so you can see what the assistant did and verify the result.

History and memory

Conversations are stored so you can resume them later. Important facts surfaced during a conversation can be captured back into the cache as chunks, making them available to future runs. Treat chat as both a working surface and a capture path: anything useful can be promoted into durable memory.