ForgeKnowledge
The engineering knowledge engine for the Forge ecosystem. Captures the why behind every automated decision — work orders, PMs, fault diagnoses, design changes — and makes it queryable in plain English.
Works fully offline
The core Q&A engine, timeline, and entity explorer run entirely on-premises. Set ANTHROPIC_API_KEY to enable richer multi-hop AI answers — opt-in, never required.
Features
Decision audit trail
Every automated decision in the Forge ecosystem — a work order created by ForgeOps, a PM generated by ForgeMaint, a fault diagnosed on a machine — is ingested via ForgeHub and stored with full context: inputs, rule applied, confidence, outcome.
Natural-language Q&A
Ask questions in plain English. "Why was work order WO-2026-0847 created?" "What triggered the last PM on VMC-01?" ForgeKnowledge retrieves the decision record and explains it in prose. Works fully offline — no cloud dependency.
Entity relationship explorer
Navigate the graph of entities and decisions. A machine links to its faults, its PMs, its toolpaths, and the design revisions triggered by its events. Explore the full causal chain from floor event to engineering change.
ForgeHub subscriber
ForgeKnowledge subscribes to ForgeHub and ingests events from every product in real time. No manual log shipping. Events arrive with full payloads and are immediately indexed for query.
Richer answers with AI
Set ANTHROPIC_API_KEY to unlock multi-hop reasoning. ForgeKnowledge uses Claude to synthesise answers that span multiple decision records — tracing a machine's degradation history across months of events, for example.
Offline-first
The core Q&A and timeline engine runs entirely locally. No data leaves your network. The AI mode (Claude) is opt-in and operates on summaries, not raw records — your manufacturing data stays on-premises.
Example queries
"Why was this work order created?"
ForgeKnowledge retrieves the ForgeHub event, the rule that fired, the sensor reading that triggered it, and the maintenance history of the affected asset.
"What changed in this design revision?"
Links the ForgeCAD cad.exported event to the ForgeOps brief that requested the change, the floor event that originated the request, and the machine it will run on.
"When did this machine last have a PM?"
Queries the ForgeMaint event history for the asset, returns the last PM date, technician, duration, and parts replaced.
"Has this part ever gone out of spec?"
Searches the full event history for quality events referencing the part number, returns a timeline of deviations with root-cause records where available.
REST + WebSocket — port 8771
Query decision records, subscribe to the knowledge stream, or ask natural-language questions via HTTP.
# Ask a natural-language question
curl -X POST http://localhost:8771/ask \
-H "Content-Type: application/json" \
-d '{"question": "Why was work order WO-2026-0847 created?"}'
# Response
{
"answer": "Work order WO-2026-0847 was created because VMC-01 reported a fixture deviation of 0.12 mm at 09:14 on 2026-05-28. The deviation exceeded the 0.08 mm threshold defined in maintenance rule MR-14 (fixture_deviation_check). ForgeMaint created the work order automatically and notified the assigned technician.",
"sources": ["forgehub.event.maintenance.work_order.created", "forgemaint.rule.MR-14"],
"confidence": 0.97
}