The context engine

FABRIC

One brain that knows the story.

Fabric connects ShotGrid, FileMaker, Airtable and the rest of your production stack into one governed brain. Ask it anything about your production: it answers in plain language, cites its sources, and only shows people what they are cleared to see.

What changes
01

Ask once, search everything

One question spans ShotGrid, the script and your DAM. Which shots used the hero prop, and is it cleared? You get the answer with versions and rights attached, not a list of places to go look.

02

Every answer shows its work

No mystery answers. Each fact carries its source and whether a human has confirmed it. Your team can trust the answer or trace it back, either way, in the open.

03

Context survives the wrap

When a show wraps, the people who knew it roll off. Fabric keeps the decisions, the relationships and the history searchable, so the next season starts from memory, not archaeology.

04

Answers respect time and clearance

Fabric knows current from superseded, so you never act on last month's truth. And permissions are enforced inside every question: no clearance, no answer.

How it does it

Native to your production stack

ShotGrid, FileMaker and Airtable connect natively, alongside DAM, Slack, git and CSV import. 300+ business systems are reachable through the connector catalog. Read-through by default: your systems of record stay authoritative, and publish-back is controlled.

One common language: MovieLabs OMC

Every source field maps to open OMC keys: 340 mapped fields across 14 entity types. AI suggests each mapping and your team approves it, so the model stays yours.

A governed knowledge graph

23 node types fuse the production world (sequences, shots, versions, assets, people, files) with the script world (scenes, cast, props, sets, locations, wardrobe and more). Duplicate entities across systems are matched and suggested, then merged only on human approval.

More than document search

Retrieval is hybrid: semantic search, structured queries, keyword and knowledge-graph traversal, merged and reranked. Answers stream with citations. Permissions filter inside the query and fail closed, and every fact is labeled trusted or proposed.

Built for people and agents

Your team gets conversational answers with sources. Your agents get 23 entity tools over MCP, the open agent-connector standard, plus organizational memory: recall, remember, forget and who_knows. The same surfaces are available over REST, including Task, Version and Note sync for production-tracking round-trips.

For your engineers

The technical depths.

Counts and limits below are verified against shipping code. The full capability datasheet is available on request.

Fabric specifications
Native connectors
ShotGrid (incl. signature-verified live webhooks) · FileMaker · Airtable · DAM (5th Kind / Backlot) · Slack · git · CSV import with column mapping; 300+ business systems reachable through the connector catalog
Live federation
Query-time federation of never-indexed sources under the caller's own credentials: Google Drive, Gmail, Google Calendar, Linear and Slack adapters, plus any custom MCP endpoint registered by URL with no code change
Data model
MovieLabs OMC: 340 mapped fields across 14 entity types, every source field mapped to open omc.* keys; AI-suggested field mappings and evidence-based relationship suggestions, human-approved
Entity graph
23 node types spanning the production world (sequences, shots, versions, assets, people, files) and the script world (scenes, cast, props, sets, locations, wardrobe and more); cross-system duplicates merged only on human approval
Retrieval
Hybrid: semantic vector search + structured queries + keyword + knowledge-graph traversal, merged and reranked; conversational answers stream with citations, session-tracked with feedback capture
Embeddings
1024-dim, computed by a locally hosted model inside the deployment environment; no calls to external AI APIs
Governance
Permissions enforced inside every query, fail-closed (no clearance, no rows); every fact carries provenance (trusted vs proposed); temporal awareness: current vs superseded, expiry, recency weighting
Organizational memory
Permission-scoped episodic session summaries (entities, decisions, outcomes; never raw transcripts; promotion-gated), preferences, forget/deletion, lineage and attribution, nightly consolidation, auto-maintained browsable wiki
Agent and API access
23 entity MCP tools plus organizational-memory tools (recall, remember, forget, who_knows) over stdio and HTTP; same surfaces over REST, including Task/Version/Note sync for production-tracking round-trips
Questions
Does Fabric replace ShotGrid or our DAM?

No. Fabric reads through to the systems you already run and keeps them authoritative. ShotGrid stays your system of record and your DAM stays the creative front door. Fabric adds one governed layer on top: normalized to MovieLabs OMC, searchable in plain language, and queryable by agents over MCP.

Does our production data leave our environment?

In production deployments, Fabric runs in your AWS account or on your on-prem hosts, and content stays in your environment. Embeddings are computed by a locally hosted model inside the deployment environment, with no calls to external AI APIs. Optional generative features can be pinned to local models or to providers you approve.

How does Fabric keep AI answers trustworthy?

Every fact carries provenance. Only deterministic structural facts are auto-trusted; everything AI-extracted waits for human promotion. Permissions are enforced inside every query and fail closed, so no clearance means no rows. Answers also respect time: Fabric knows current from superseded and weights recency.