Restrict agents by plant, production line, product, supplier, and engineering program.
Industrial boundaries become access boundaries — an agent for one line, plant, or program ends exactly there.
Govern agents connecting engineering, production, quality, maintenance, and supply-chain systems.
Manufacturing agents may access product designs, supplier records, production schedules, quality systems, maintenance platforms, and operational technology. A mistaken action can affect not only data but physical production, worker safety, and product quality.
Example hierarchy. The plants, lines, assets and release rights are the ones you set.
Industrial boundaries become access boundaries — an agent for one line, plant, or program ends exactly there.
Telemetry and schedules are readable without any path to altering setpoints, recipes, or sequences.
Actions that can stop a line, affect quality or endanger people are not in the tool list an agent receives, so they stay with qualified staff.
Design and process data carries its classification, and access follows program authorization — including export-control rules.
Supplier communication stays inside the purchase order and vendor relationship it concerns, and it goes out as the buyer the agent acts for — never beyond that person’s authority.
Quality investigations walk the complete chain — alert, analysis, action, verdict — in one record.
Explore the context attached to every request.
maintenance-agent has a defined purpose and a bounded set of tools.
The request acts for plant engineer, using that person’s permissions.
The policy applies to the requested action within Plant systems, before the tool executes.
Maintenance agents read sensor data and open work orders; changes to equipment remain with technicians.
Investigation agents assemble the production history around a defect without the ability to alter the records they analyze.
Planning agents propose schedules within their plant and line scope; committing a change is not an action they hold, so a planner does it.
Onboarding agents collect documents and draft records; activation and payment terms are not actions they hold, so procurement sets both.
Retrieval runs within the requester’s program authorization, keeping export-controlled designs contained.
Assistants answer from released, current instructions — never drafts or superseded revisions.
No. Fabriq drops in in front of the IT and OT systems agents call — nothing about the equipment or platforms changes. Read access and change authority are separated per request, so analytics never quietly becomes control.
Not unless you grant it that. Read access to operational data is a separate action from changing a setting, and safety-, quality- and production-impacting actions can be left out of every agent’s grant — in which case they never appear in a tool list at all.
Who asked, which agent acted, what it touched, and the verdict — one trail across engineering, planning, and quality systems.
Yes — per user. An engineer’s agent reaches the designs and process data of the programs that engineer is authorized for, including export-control boundaries, and an out-of-authorization retrieval fails at the access layer and is logged.
Over an outbound-only encrypted tunnel — no inbound ports to open — with OpenAPI import for internal APIs and read-only guardrails for Postgres.
Yes — disable it org-wide and its next call fails closed; sessions are visible live and revocable now, not at token expiry.