The hive
A new chat starts by loading the hive for its project: the rules, what's open, and what changed last time. A chat wraps itself up when you start another, close the window or leave it for 55 minutes: it logs the work and saves a short summary. Wrap up does the same on the spot. A clean wrap ends with Cells interlinked.: safe to leave. No line? Read what's pending first.
Short-term memory
For things that matter for days, not forever. The assistant adds them herself. Each has a reach (everywhere, one project, a type of project, one mode) and ends by itself: on a date, after a few chats, or when a task is done. Open chats hear about new ones with your next message. Extend, kill, pin or promote one to a rule any time.
What it knows about you
Set your name, full name and nicknames, and how freely it mixes them. What you tell the assistant about yourself (health, family, tastes, how you like to work) it saves on its own, in your own words. Up to 10 pinned facts show in every chat; the rest it recalls when they matter. Ask it to pin, change or forget any of them.
The attic
Before each message, Interlink searches the hive by meaning (a small model on your PC, no internet) and by words, and quietly adds up to 3 old notes that fit what you said. Notes that keep showing up unused stop coming. The assistant tags notes with other words you might use, so they're easy to find again.
Rules
Rules opens full screen. Every rule shows its scope, category, how many tokens it costs per chat and its age. Sort, filter, bulk-retire, rescope, edit. Short forms keep every chat start cheap, and On demand rules load only before the work they're about. A Start cost tab shows what a chat start costs, part by part.
Add a rule
When the assistant suggests a rule it comes with an Add rule button. It opens the editor filled in: pick who it's for and a category, then Save. Nothing's added until you save. If it says much the same as a rule you already have, Save offers to merge them.
Lessons become rules
Lessons the assistant logs come back when they matter: editing a file brings the lessons about that file along. Once a day, the same mistake logged again and again becomes a rule, with a toast to OK or VETO it.
Context depth
How much a new chat brings along, in three levels: Tears in Rain (least), Interlinked and Replicant (most). More context means fewer wait, what were we doing? moments; less means cheaper chats. Each level shows its cost.