Marble philosopher bust dissolving into a temporal company knowledge graph

First of a kind / Context handoff API

FOAK

Give agents judgment before action.

FOAK turns scattered company artifacts into temporal relationships, then lets retrieval agents follow the trail before the work begins.

Book a callcontext_handoff(task)

Living knowledge graph

Company memory needs structure, sequence, and judgment.

FOAK stands for first of a kind: an agent-native memory layer where company knowledge is alive enough to be traversed, not merely searched.

The current failure mode

Agents are smart enough to work, but not organized enough to know what your company knows.

Primary agent

GmailNotionGitHubSlackCalendarLinear

reads too muchmisses relationshipsbloats the context windowproduces weaker output

Context handoff

Company memory should be relational by default.

Vector search finds similar text. FOAK follows who, what, when, and why across people, projects, meetings, PRs, contracts, and decisions.

await foak.context_handoff({
  task: "Prepare everything needed to hire Theo to work on FOAK."
})

taskentrypointsgraph traversalcontext pack

How it works

FOAK makes retrieval behave like a sharp internal operator.

Connect company data

Slack, Notion, Gmail, Calendar, GitHub, Linear, CRM, meetings, contracts, and docs become typed memory.

Build the living graph

People, projects, artifacts, decisions, commitments, and events are connected through time.

Traverse relationships

Specialized retrieval agents inspect compact graph cards and follow only the relevant paths.

Return context packs

FOAK materializes cleaned threads, notes, summaries, source handles, and the facts needed to act.

Concrete scenario

Prepare everything needed to hire Theo for FOAK.

Task

"I am about to hire Theo. Pull together everything relevant before I draft the contract."

Theo

candidate, collaborator, prior PR

Email thread

scope, rate, availability

Calendar call

decision notes and open terms

GitHub PR

proof of work on FOAK

Contractor policy

approval path and clauses

Context pack

Relevant email thread

Call notes

PR context

Project plan

Policy summary

Source handles

Why now

Agents are becoming the interface to work. Company context is still trapped across tools.

MCP gives agents access, but not judgment about what context matters.

Context windows are larger, but irrelevant context still makes agents slower and less reliable.

Cheaper inference makes specialized retrieval agents practical.

Different by design

Task-first, graph-based, temporal, and context-minimizing.

Existing thingLimitationFOAK

MCP toolsAccess to every source, but the primary agent still has to search each one.One task handoff, with retrieval handled by graph agents.

Vector searchSimilar chunks without durable people, project, decision, and time relationships.Temporal graph traversal before any rich payload is loaded.

Enterprise searchDesigned for humans typing queries and choosing links.Designed for agents completing tasks.

Manual docsRequires humans to keep the map clean.Turns working artifacts into connected memory.

First of a kind

Let agents work from memory, not guesswork.

Book a call