Log in

Knowledge Graph

Your org's context — playbooks, decisions, customer history, what's true today — kept fresh and accessible to every workflow.

The Knowledge Graph is your team's shared context, made first-class. Playbooks, product decisions, customer history, internal policies, what's true today — all stored in one place, updated continuously, and pulled into every Skill the team runs.

Without it, a workflow that depends on "the way we do things" only works for the person who already knows. With it, the way you do things becomes a teammate any skill can ask.

What's In It

The Knowledge Graph holds whatever your team needs to do its work consistently. Common shapes:

  • Playbooks — How we run a launch. How we triage a P1. How we respond to a renewal-at-risk signal.
  • Decisions — What we chose, why, and what we'd revisit.
  • Customer context — The account history, the active deals, the open commitments, the dropped balls.
  • Product truth — What shipped, what's in flight, what's deprecated.
  • Policies and standards — Brand voice, security posture, legal red lines, escalation thresholds.

Anything that would normally live in five different Notion docs, three Slack threads, and one person's head.

Curated, Not Crawled

The Knowledge Graph is deliberately not a crawl-everything index. Random ingestion produces a search bar; that's not the goal. The goal is a small, trusted, owned body of context that any Skill or agent can rely on without checking.

Entries are added in two ways:

  1. Direct authoring — A teammate writes or imports a piece of context (a playbook, a decision record, a brand standard).
  2. Promoted from work — A Shared Conversation or finished task surfaces something worth keeping. A teammate promotes it into the graph with one click.

Either path runs through review before it becomes published context. Nothing lands silently.

Always Fresh, Never Drifting

Context decays. A playbook from six months ago might describe a stack that's since been ripped out. The Knowledge Graph handles this two ways:

  • Source attribution — Every entry shows what it came from, when, and who's responsible for it.
  • Review nudges — Entries past a freshness threshold surface for re-review. Stale items can be archived without losing history.

When a Skill cites a graph entry that's been marked stale, the Output Ledger flags it on the resulting draft. Reviewers see the risk before they approve.

How Skills Use It

When a Skill runs, it pulls from the graph automatically. A "Q3 Launch Brief" Skill might draw on:

  • The launch playbook
  • The current quarter's GTM template
  • The product release notes (linked from Jira / Linear)
  • The brand voice standard
  • The last three launch briefs and their review outcomes

No teammate has to remember to feed the agent any of this. The Skill declares what it needs from the graph; the graph provides; the Agent Workspace drafts.

Why It's Foundational

Most "shared AI" experiments fail at the context layer. Each user re-explains the company to the model. The agent acts confidently on stale or invented facts. Outputs disagree across teammates because each is working from a different mental snapshot.

The Knowledge Graph fixes that by giving the team one source of truth for how we do things here — and making sure every skill the team uses pulls from it.