Where enterprise intelligence goes to work.
People, knowledge, systems, and AI agents - connected in one governed workspace.
- Agentic conversations featuring custom tools and office file integration
- Personalized workspace tools and enterprise knowledge base integrations
- Enterprise-grade security and governance to transform static knowledge into actionable decisions
Welcome to KEOS
Sign in with your enterprise account to begin your personalized onboarding.
The AI operator that stops refinery margin leaking between shifts.
Refinery.AI watches operating conditions, spots drift, explains what matters and recommends the next move. The operator stays in control.
From signal to signed-off dollars
Reads pressure, flow, temperature and valve position from the historian. Read-only.
Names the likely cause, shows the evidence and prices the drift in dollars per day.
Tests the move on the twin first. The operator approves or declines with a reason.
Measures against a locked 30-day baseline. Only finance-signed dollars count.
Same drift. Two different mornings.
Without Refinery.AI
- 03:07Air damper drifts. Nothing alarms yet.
- 03:40Alarms arrive in a burst. The board operator is mid-shift handling three other things.
- 04:30Someone connects the alarms to F-101.
- 06:00Handover notes say "watch F-101". The cause is still unknown.
- 08:00About 5 hours drifted. At $1,200 an hour that is about $6,000.
With Refinery.AI
- 03:07Same drift begins.
- 03:12Refinery.AI flags it before any alarm and names the air damper.
- 03:14It shows the evidence, the price per day and the move.
- 03:20The operator approves after the twin test passes.
- 03:35Back on target. About 28 minutes drifted, roughly $560. The full story is in the handover.
90 days. Your data. Your number.
Backtest for 30 days, advise for 30, measure for 30. No live changes until you say so.
Choose an agent
One shows the plant and the money. The other tracks emissions against permit limits.
Your margin agent
Noor prices every drift on CDU-1 in dollars: loss, profit and opportunity. Illustrative data.What Noor sees, by equipment (USD k/day, tap a bar to open it)
Emission agent: Motiva Port Arthur, Texas
Aerial methane survey for the Port Arthur site, in the Bridger-style data format.
Sample data placed around the Port Arthur, Texas refinery site. Detections, positions, rules and pictures are invented: not Bridger data, not refinery data, not legal advice.
Agent trace (tool calls)
- Tool calls appear here.
Survey data
Tap a row to see every field, grouped as in the data specification, with the plume map and aerial view. Units: ppm, SCFH (standard cubic feet per hour), feet.
Audit log
| Time | Mode | Action | Location | Outcome | Why |
|---|
Nothing yet.
Permissions, rules and assumptions
What each mode may do, and what the agent may never do.
| Action | Risk | Assist | Advise | Autonomous |
|---|
Priority score = size (under 100 SCFH is 1, 100 to 499 is 2, 500 or more is 3) + 1 if persistent + 1 if the rate rose 25% or more from the first detection to the latest + 1 for flares and compressors. Score 5 or more is High, 3 to 4 is Medium, otherwise Low. All thresholds are invented. 500 SCFH is roughly 10 kg/h, assuming about 0.0192 kg of methane per standard cubic foot. “Persistent” is taken as given in the data (here: detected in 2 or more consecutive scans); confirm Bridger’s own definition. Flare and compressor detections can come from normal operation, so check operating logs before treating them as leaks. The plume map is a simulation drawn from each row’s values, and the map is drawn from coordinates around the example site with invented positions; neither is real imagery. Real Bridger plume maps and aerial photos would replace them when a real data feed is connected.
Industrial Security Agent
Built on the Refinery.AI Governance FrameworkAutonomy level
Pending approvals
Governance watchlist
Top 10 priorities: starting importance score and owner.
Audit trail
- No actions yet.