Concepts
Six words carry most of Flipper AI. Learn them here and the rest of the documentation reads quickly.
Episode
A raw thing that happened: a Slack message, an email thread, a document revision, a calendar event. Flipper AI is a facts database, not a document store — it does not keep your documents. What it keeps is the short verbatim quote each fact was extracted from, encrypted, so every fact can show the words it came from.
Fact
One claim, written as subject – predicate – object: *Payments API v2 — blocked by — the ledger schema review*. Facts carry a source, a timestamp, a trust score and a sensitivity. They are never edited in place.
Entity
The subject a fact hangs off: a person, project, customer or team. Entities carry aliases, so @sam, sam@kepler.dev and “Sam Okafor” all resolve to one node in the graph.
Trust score
Every fact carries a trust score from 0 to 100, and hovering it shows the four things that produced it: whether the claim was found quoted in its source, how many independent sources assert it, whether anything live contradicts it, and how much it has aged — faster for a status than for a reporting line. A fact ingested before Flipper AI started quoting its sources shows a grey score and says so, because it cannot be re-verified.
Supersession
When something changes, the old fact is not deleted — it is marked superseded and points at its replacement. get_fact_history walks that chain, which is how “we moved the launch to Q4” stays legible six months later.
Playbook
A repeatable process — refunds, escalations, onboarding — distilled from your own facts into steps an agent can follow. A playbook is inert until a human approves it. See Playbooks.