Coat & Calender agent
Controls slurry mixing, slot-die coating, drying and calendering for loading, coat weight, thickness and porosity uniformity across the full web width.
Each Anodela agent owns a step of cell manufacturing, runs at the factory edge, and hands off through a shared model of the electrode, the jelly roll and the cell.
Agents in production across 14 lines [PLACEHOLDER]
Controls slurry mixing, slot-die coating, drying and calendering for loading, coat weight, thickness and porosity uniformity across the full web width.
Senses and predicts coating defects, wrinkles, pinholes, streaks, burrs, metal particles, weld quality and internal defects from vision and inline X-ray/CT.
Controls stacking and winding, tab and terminal welding, electrolyte fill and sealing to sub-millimetre tolerance with void and misalignment prediction.
Optimizes formation cycling, aging, degassing, end-of-line test and grading/binning to cut hours, energy and working capital per cell.
Balances the moving line, optimizes first-pass yield, scrap recovery, WIP routing and takt when a bottleneck or quality risk appears.
Drives robots, AGVs and dry-room material handling for electrodes, jelly rolls and cells with fail-safe stop behaviour.
Optimizes right-first-time, non-conformance handling, cell genealogy and traceability evidence under IATF 16949 and IEC 62660.
Simulates coating, drying, calendering, assembly and formation as-built versus as-designed and hits target yield, quality and takt before the run.
An agent is not a chat window. It is a bounded control loop with grounded perception, an approved action space, a simulation gate and an immutable record.
Every inference cites the sensor stream, the recipe revision and the standard clause it was evaluated against. Engineers can trace any decision to its evidence in one click.
Each agent can only act inside an envelope your quality team approved and the twin validated. Everything else is a proposal that waits for a human.
When an engineer rejects or edits a proposal, that becomes training signal. The action-result history — proposal, approval, executed change, rollback, downstream outcome — is far harder to copy than labelled images alone.
All agents read one fused representation of the electrode, jelly roll and cell — no duplicated or conflicting perception.
When quality risk and takt conflict, safety and cell quality win; the orchestrator resolves and records why.
A formation anomaly can be traced back to a coating zone, a slitting burr or a weld station via cell genealogy.
Anything outside the envelope, novel or safety-relevant escalates to the named engineer with full context.
Shadow, advisory, then bounded control — each stage unlocked by measured accuracy against your own engineers and validated in the twin.
Before an agent is allowed to act, its envelope is proven in simulation against as-built line behaviour — not the as-designed drawing.
Simulate thousands of recipe and takt scenarios per line: coating and drying, calendering, stacking and winding, welding, fill, sealing, formation scheduling and robot flow. Commission new lines and new chemistries with the window already found.
The twin is continuously reconciled with real line telemetry.
Every validated envelope becomes reusable qualification evidence.
Predict throughput impact before changing a single setpoint.
Move a proven window from NMC to LFP or silicon-anode faster.
“Formation was 24% of our cell cost. Anodela re-sequenced channels and trimmed the protocol per cell, and we took 17% of the formation hours out without touching a single capacity spec.”
“Metal-particle escapes were our nightmare. The contamination model flags a signature our inspection vendor simply does not see, and it quarantines the affected cells automatically.”
“We commissioned line four with the twin first. It predicted the calendering window that would hold porosity, and the real line matched it within tolerance on week two.”
Each level has entry criteria, a measured accuracy bar and an exit path back to the previous level if performance degrades.
| Level | What the agent does | Human role | Entry criteria |
|---|---|---|---|
| L0 — Shadow | Observes and predicts silently | Full control | Connectors live, baseline captured |
| L1 — Advisory | Recommends moves with citations | Approves every action | Accuracy at or above engineer baseline |
| L2 — Bounded control | Executes inside an approved envelope | Supervises exceptions | Twin-validated envelope, quality sign-off |
| L3 — Supervised autonomy | Runs the step end to end | Reviews exceptions and outcomes | Sustained accuracy plus audited ROI |
Cell manufacturing is safety-critical and IP-dense. Anodela is designed so quality, IT/OT and EHS approvers can say yes.
SOC 2 Type II program with encryption in transit and at rest, SSO/SAML, SCIM and RBAC down to the line.
Chemistry, coat-weight and formation recipes stay in your tenant. Optional fully on-prem deployment with no egress.
Every perception, proposal, approval and executed setpoint change is immutably logged for IATF 16949 and IEC 62660 evidence.
EU, US, Korea, Japan and China-domestic regions, or air-gapped factory edge for sovereign programs.
Pre-built OT and API connectors read process, vision, X-ray/CT, formation and MES data — and write bounded setpoints back. Nothing rips and replaces; Anodela runs alongside the equipment you have qualified.
They coordinate through the orchestrator and a shared world model rather than free-form messaging, so every hand-off is typed, bounded and auditable.
Fine-tuned vision and time-series models for perception and control, served on-site; frontier models from Anthropic and Google for battery-process and safety reasoning. A router picks the best and cheapest model per step.
Yes. Agents are independently licensed, armed and disarmed per line, and any agent can be returned to advisory or shadow instantly without stopping production.
Pick the workflow that costs you the most scrap. We will run it in shadow and show you the gap.