Docs/ Rules, context & knowledge

Context engine (client brain)

Per-client profiles that auto-load when you reply, so drafts use real facts.

The context engine gives the AI agent a structured memory for each client — so every draft reflects your actual relationship, not a generic guess.

What a profile contains

A context profile is attached to a client domain (e.g., acmecorp.com). Every field you fill in becomes a source of truth the agent draws on when it drafts a reply, generates a follow-up, or checks commitments before you send.

Example profile card — Acme Corp (acmecorp.com)
StatusActive client · renewal due 2026-09-01
Key contactsMaya Patel (Champion), Derek Osei (Legal)
ToneFormal; avoid casual openers; use 'Warm regards'
Open loopsAwaiting countersigned SLA (sent 2026-06-01)
Don't forgetDerek is on leave until 2026-07-15; copy Maya instead

Domain matching

A profile activates whenever you open a reply to any address @that-domain. You can also pin a profile to specific addresses rather than the whole domain.

Auto-load on reply

The moment you open a compose or reply window addressed to a profiled domain, the agent silently loads that profile into its working context. You don't press a button — the profile just appears in the sidebar panel so you can see exactly what the agent is reading.

The agent uses the profile's tone rules, open loops, and 'don't forget' notes when generating its first draft. If a value is not in the profile, it will ask you rather than invent something.

No hallucinated facts

The agent is instructed to use only values present in the active profile (or resolved variables). It will flag a gap rather than fill it with invented details.

Commitment tracking

Before you send, the agent scans both your outgoing draft and the active profile's open loops. If it finds an unaddressed commitment — a deliverable you promised that isn't mentioned in the draft — it surfaces a warning in the send confirmation panel.

You can dismiss the warning, add a note, or mark the loop closed. Closed loops are kept in the profile history so you have a record.

  1. 1

    Add an open loop

    In the profile editor, open the 'Open loops' section and add a plain-text commitment with an optional due date. The agent will surface this on every reply to that domain until you mark it closed.

  2. 2

    Review before sending

    The send confirmation panel lists any open loops the current draft doesn't address. Click a loop to expand it and decide: dismiss, draft a line about it, or mark it closed.

  3. 3

    Close a loop

    Mark a loop closed from either the profile editor or the send panel. Closed loops move to the profile's history tab; they no longer trigger warnings.

Context folders

When you have many related clients — a portfolio of accounts, a cohort of job applicants, a roster of vendors — group their profiles into a context folder. Each folder has a purpose statement the agent reads as additional framing before it loads an individual profile.

For example, a folder named 'Enterprise accounts' with the purpose 'These are high-touch customers with SLAs; always be formal and never commit to timelines without checking with the team first' provides policy context that applies to every profile inside it.

Use folder purpose for shared policy

Put rules that apply to an entire category of client in the folder purpose, so you don't have to repeat them in every individual profile.

Profile variables

Any field in a profile can be referenced as a {{variable}} in templates, rules, and AI prompts. Values defined on the profile take precedence over folder-level and global variables when there is a naming conflict. See the Variables doc for the full precedence order.

Frequently asked

Feature overview

Rules & Brain

Ready to try it?

Start free