Knowledge Intelligence

What Knowledge Intelligence Actually Is

Knowledge Intelligence is built around one primary artifact: the meeting transcript. Not code knowledge graphs. Not wiki pages. Meetings — the place where decisions actually get made and then immediately forgotten.

The core flow is straightforward:

• → A meeting recording tool (like Fathom) records your meeting
• → The provider sends a webhook to InteliG with the transcript, participants, and summary
• → InteliG ingests it as a Knowledge Artifact with a full Transcript Aggregate
• → AI automatically extracts structured intelligence: decisions, action items, key topics, sentiment, and an executive summary

The dashboard tracks three numbers: artifacts (meeting count), decisions (extracted), and action items (pending). That's it. No vanity metrics.

How It Works

The processing pipeline is fully async. Webhooks are received fast; intelligence extraction happens in the background.

Integrations

Knowledge Intelligence uses a provider adapter pattern. Connect a meeting recording tool, and InteliG handles the rest.

Every integration follows the same lifecycle:

• → DISCONNECTED — No connection. Starting state.
• → CONNECTING — API key validated, webhook URL generated.
• → BACKFILLING — Importing historical meetings from the provider.
• → ACTIVE — Receiving webhooks in real time. Health metrics tracked: last webhook timestamp, 24h webhook count, total meetings ingested.

Artifacts

Every meeting becomes a Knowledge Artifact. Each artifact stores:

• → Provider and type — Where it came from (Fathom, manual, etc.) and what it is (meeting transcript, voice recording).
• → Title and participants — Meeting title (AI can override; original preserved), attendee names, emails, roles, and linked contributor IDs.
• → AI summary — Executive summary, key topics, and sentiment analysis.
• → Intelligence counts — Decision count, action item count, and a flag indicating whether intelligence extraction has completed.
• → Scope — Organization-wide or personal visibility, with owner tracking.

Beyond automated ingestion, you can also create artifacts manually (paste a transcript and give it a title) or via voice recordings with contexts like thinking out loud, decision capture, and idea brainstorm. All go through the same intelligence extraction pipeline.

Decisions

Decisions are extracted from transcripts by AI. The extraction is deliberately strict: only explicit commitments count. "We will ship the new API by Friday." "Let's go with Postgres." "Agreed to drop the mobile app from the roadmap." Suggestions, considerations, and maybes are not decisions.

Each decision carries:

• → Content and context — The decision statement plus the surrounding discussion that led to it.
• → Decided by — The person (or "group") who made the call.
• → Confidence — HIGH, MEDIUM, or LOW, based on how explicit the commitment was in the transcript.
• → Source — PROVIDER (from the recording tool), AI (extracted by InteliG), or MANUAL (added by a user).

Quality over quantity. A meeting with zero extracted decisions is better than one with five weak ones.

Action Items

Action items are extracted alongside decisions. The AI requires clear ownership — items without an identifiable assignee are skipped entirely.

• → Content and assignee — What needs to be done and who's responsible. Linked to InteliG's contributor identity system for cross-pillar tracking.
• → Due date — Captured from deadline indicators in the transcript ("by Friday", "next sprint", "end of quarter").
• → Status tracking — PENDING, IN_PROGRESS, COMPLETED, or CANCELLED. Status transitions are validated — you can't reopen a completed item.
• → Overdue detection — Built in. If an action item has a due date in the past and isn't in a terminal state, it's flagged.

Signal — Syncing to Git

Knowledge artifacts can be synced to GitHub as structured markdown files via InteliG's Signal system. Only organization-scoped artifacts are eligible for auto-sync.

Each synced meeting produces a markdown file containing:

• → Meeting metadata (date, type, source link)
• → Executive summary
• → Decisions (bulleted, with who decided)
• → Action items (table: action, assignee, due date, status)
• → Participants list and full transcript

Commit message format: chore(signal): sync meeting — { title }. This makes meeting intelligence a first-class citizen in your repository history.

How Knowledge Feeds Cognis

Knowledge doesn't exist in a vacuum. Everything extracted from meetings flows into Cognis, InteliG's AI reasoning engine:

• → Domain events — Every ingested artifact publishes a KnowledgeIngestedEvent to Kafka, making it available to all bounded contexts including Cognis.
• → Queryable intelligence — Structured outputs (decisions, action items, topics, sentiment) are persisted and queryable. Cognis can reason across all meetings.
• → Contributor linking — Action items link to contributors, connecting meeting intelligence to InteliG's identity graph. "What action items are assigned to this person across all meetings?" becomes a simple query.

Without sources, decisions stay buried in recordings. With sources, Cognis surfaces what matters. Ask it:

Use with Agents

One agent playbook handles knowledge onboarding end-to-end:

See all playbooks →

Related

• → Cognis AI Engine — The reasoning layer that synthesizes knowledge intelligence with all other pillars.
• → Strategy Intelligence — Connect meeting decisions to strategic alignment tracking.