Embedded systems agents for commodity producers
Turn plant data into work your agents can execute.
Import Intel connects PLCs, sensors, optimizers, legacy systems, business data, and commodity-market context so your team can understand production and delegate operational work in plain language.
Read-only by default. Source-grounded. Human-approved where actions matter.
Import Intel · operations platform
Illustrative composite
Ask
“Compare output and downtime by machine for the latest seven operating days. What is the largest verified constraint?”
Agent activity
- Schema inspectionApproved tables and columns only
- Semantic profileMachines, metrics, units, shifts resolved
- Read-only queryOne bounded SELECT · statement timeout
- ComparisonOutput and downtime compared by asset
Governed dashboard
Values redacted
Output and downtime by machine
- Machine center A
- Machine center B
- Machine center C
- Machine center D
Source: approved production and downtime facts · Time scope: latest seven operating days · Returned rows and generated query inspectable in product
Approval gate
Ranked action plan prepared. Purchase-order draft is review-only; nothing is sent, posted, or issued without a person.
- 01
Plant, mill, and office systems
PLCs, sensors, optimizers, production databases, ERP, documents, email
- 02
Import Intel Edge and connectors
Local collection, durable queueing, retried uploads — no inbound connection into plant systems
- 03
Facts, ontology, and governed knowledge
Raw-first ingest, deterministic fact extraction, shared operational model, saved knowledge
- 04
Agents, skills, and tools
Reusable roles, digital standard operating procedures, governed tool access
- 05
Answers, dashboards, reports, and draft actions
With source evidence, time scope, and approval gates attached
Source types mapped during deployment
- PLC + machine buffers
- Sensors
- Optimizers
- Downtime systems
- Production SQL
- ERP adapters
- Work orders
- Parts + stock
- PDFs + daily reports
- Microsoft 365
- Public market data
What the platform guarantees
01
Grounded in plant data
Answers come from your connected sources, not from model recall.
02
Governed read-only queries
One bounded SELECT, blocked mutations, statement timeouts.
03
Tenant-scoped access
One company's plant data is never pooled into another's context.
04
Source and freshness visible
Source system, returned rows, and time scope stay attached.
05
Review before consequential actions
Sends, posts, purchases, and stock changes wait for a person.
The operating reality
Your plant already generates the data. It does not have one operating layer.
The data is fragmented
PLCs, optimizers, production databases, ERP, documents, and spreadsheets each describe a different part of the same operation.
The meaning is inconsistent
A machine name, source key, metric, shift, or downtime reason can mean different things in different systems.
The work after the answer is still manual
Someone still has to build the report, coordinate the incident, draft the follow-up, or prepare the purchasing request.
Question to verified action
One question in. Evidence, output, and a review boundary out.
01 · Sources selected
- Production facts from connected machine sources
- Detected and classified downtime events
- Site ontology: machines, metrics, units, shifts
02 · Checks performed
- 01Resolve “slowed us down” into output and downtime concepts
- 02Confirm which sources and time ranges actually contain data
- 03Run tenant-scoped, read-only queries over approved facts
- 04Quantify the constraint from the rows that were returned
03 · Output
One verified constraint, the evidence behind it, and three ranked draft actions with owners and verification metrics.
04 · Review boundary
The plan is a recommendation. Nothing is changed in any plant or business system.
Human approval required
Featured workflows
One platform for the work that crosses the plant.
Live plant command center
Plant or mill manager
Ask · What a person types
Build my morning dashboard for production and downtime.
Inspect · Systems read
- PLC and machine data via Import Intel Edge
- Optimizer and production databases
- Downtime and machine-event sources
Produce · Work performed
- 01Read current or latest-available production facts for the requested scope.
- 02Build the production trend and compare output by machine and shift.
- 03Build the downtime trend and compare downtime by machine.
- 04Keep source, returned rows, and time scope attached to every tile.
Review · Where it stops
A governed dashboard the manager can inspect or edit tile by tile, with the query, returned rows, and time scope behind each view.
Boundary: Live correctness depends on configured source connections and current edge uploads. Empty results stay empty — sample data is never substituted for a live query.
Industries
One platform for base-commodity production.
Representative workflows — same platform architecture
Lumber and wood products
Sawmill and planer production, optimizer data, machine downtime, grades, inventory, and maintenance across mills with a mix of modern and legacy systems.
Representative workflows — same platform architecture
Aggregates and concrete materials
Crushers, screens, conveyors, and batch processes, where throughput, stockpiles, quality, and regional demand drive the operating decisions.
Representative workflows — same platform architecture
Metals and recycling
Processing lines where equipment utilization, material grades, recovery, inventory, and commodity pricing all sit in different systems.
Representative workflows — same platform architecture
Petrochemicals and chemicals
PLC-heavy continuous or batch processes where throughput, quality, maintenance, inventory, and market context have to be reconciled across systems.
Product surfaces
More than chat. One connected operating workspace.
Conversations
Ask once. The platform finds the right systems, evidence, and workflow.
- Ask operational questions in plain language
- Query governed plant data and attach files
- See agent tool activity, sources, and returned rows
- Receive artifacts such as reports and PDFs
- Delegate work to a specialized agent and continue later
Dashboarding
The chart, answer, rows, query, source, and time scope stay connected.
- Create one chart or a full dashboard from a normal request
- Edit a selected chart conversationally
- Inspect the governed query and the rows it returned
- Keep time scope and source freshness visible
- Surface no-data and query errors honestly
- Pin results from a conversation and save reusable views
Agents
Build a repeatable role, not another one-off prompt.
- Define the job, tools, sources, and permissions for a role
- Attach files, knowledge, and skills
- Set instructions and approval rules
- Reuse the role across shifts, sites, and teams
Integrations and tools
Governed access to the systems that already run the plant and the office.
- Operational database context and read-only queries
- Production, downtime, and dashboard metrics
- Document classification, extraction, and report comparison
- Maintenance parts, inventory, order, and invoice workflows
- Market research with explicit sources
- Connected-system tools that run with the signed-in user's permissions
Knowledge and modeling
Model the operation once. Reuse the meaning across questions, dashboards, and agents.
Ontology
Import Intel learns the structure of the operation, not just the names of its database columns.
Organizations and permissions
One company's plant data is not pooled into another company's operating context.
Files and artifacts
Agent work produces things people can use, not only text.
How it works
Connect the operation. Give agents the context to work.
Connect — Connect the operation
Bring in approved plant, business, document, and market sources — without opening an inbound path into plant systems.
Understand — Build operational context
Map machines, metrics, parts, work orders, customers, and workflows into one operational model so employee language resolves to the right facts.
Govern — Enforce the boundaries
Agents interpret intent. The platform decides which data, tools, queries, and actions are allowed.
Execute — Deliver useful operational work
Answers, dashboards, reports, and review-ready actions a non-developer can use.
Architecture stack
- 01
Plant, mill, and office systems
PLCs, sensors, optimizers, production databases, ERP, documents, email
- 02
Import Intel Edge and connectors
Local collection, durable queueing, retried uploads — no inbound connection into plant systems
- 03
Facts, ontology, and governed knowledge
Raw-first ingest, deterministic fact extraction, shared operational model, saved knowledge
- 04
Agents, skills, and tools
Reusable roles, digital standard operating procedures, governed tool access
- 05
Answers, dashboards, reports, and draft actions
With source evidence, time scope, and approval gates attached
Governance
Agents can move fast without operating outside the lines.
Approved sources only
Agents read the schemas, tables, and columns a deployment approves — not whatever they can reach.
Tenant-scoped permissions
Data, dashboard sessions, artifacts, and saved knowledge stay inside the organization and the user's role.
Governed read-only queries
Read-only credentials, one bounded SELECT, blocked mutations, statement timeouts, bounded row counts.
Human approval before it matters
Messages stay unsent, invoices unposted, purchase orders unissued, inventory unchanged, machine controls untouched.
Integrations
Connect what the plant already uses.
Plant and machine data
PLCs, sensors, machine buffers, optimizers, and control-system sources collected through Import Intel Edge.
Integration-ready — source mapping is part of deployment.
Operational databases
Read-only Postgres and other operational databases holding production and process history.
Available when configured with read-only credentials.
ERP and accounting
Order, invoice, and accounts-receivable review through adapter contracts.
Adapter contracts today; live adapters require validation. No posting.
Maintenance and inventory
Work orders, assets, maintenance parts, stock balances, and locations.
Read-only and review-only. No reservation or deduction.
Documents and reports
PDFs, Word files, spreadsheets, and system exports with text and field extraction.
Available; storage and permission design validated per deployment.
Microsoft 365
Outlook, Teams, SharePoint, and OneDrive search and review workflows.
Requires configured and validated delegated tenant access.
Public and market data
Official, licensed, or current public commodity and market sources with citations.
Advisory research. No unlicensed premium feed access.
Why we pivoted
We started with commodity data. The harder problem was inside the plant.
So we rebuilt the company around that problem — an agentic operating layer for commodity producers, where the plant's own evidence drives the answer and a person still approves anything consequential.
Start with one operating question your plant cannot answer fast enough.
We will map the workflow, connect the minimum required sources, and show what an agent can complete with your data and your approval rules.