Getting started

What is Flipper

Flipper reads the places your team already works — Slack, Gmail, Drive, Calendar, Notion, Jira, Trello — and turns what happened there into facts your AI agents can look up. It is the memory layer, not another chat window.

The problem it solves

Every agent you use starts from zero. You re-explain the project, the customer, the decision you made last Tuesday, and the agent still answers like a stranger who joined this morning. The context exists — it is just scattered across five tools and nobody's head.

Flipper ingests those sources continuously, distils them into subject–predicate–object facts with confidence and provenance, and serves them over MCP. Any MCP-capable agent — Claude Code, Cursor, Codex, Claude.ai, ChatGPT — calls get_context and starts the conversation already knowing your team.

Home is the daily surface: what changed, what is waiting on you, what is stuck.

What it is not

Not a chatbot. Flipper has no chat interface of its own. Your agent is the interface.
Not a search box bolted onto your files. Facts are extracted and versioned, so a superseded decision is visibly superseded rather than returned alongside its replacement.
Not an autonomous actor. Flipper drafts with propose_action; your agent executes under its own credentials. Flipper never sends an email or posts a message.

The shape of the system

LayerWhat it does

ConnectorsPoll or receive events from your sources and store raw episodes.

SynthesisAn LLM turns each episode into facts with a confidence band and typed edges.

GraphFacts hang off entities — people, projects, customers — with aliases resolved.

GovernanceRoles, groups and sensitivity decide who and which agent sees which fact.

MCP serverTwo dozen tools that expose all of it to your agents, rate-limited and audited.