Every model, prompt and policy that can touch a line carries a version, and the run record names the versions that produced it.
Changelog
Every release that can touch a line, written down.
Nothing has shipped to a plant yet. This is the release line the first design partners will see, in the order it is planned, plus the policy that governs how any release reaches a line.
Illustrative scenario: Plant run · one architectural deep tint base from dosing to pallet · bead mill, let-down tank, tint station, filling line. Not a measured run.
Releases
The planned release line
Status is honest: everything is queued. Order follows the first-release plan.
-
v0.1Connectors and shadow mode QUEUED
Connectors to the wedge line's mill or dispenser PLC, spectrophotometer, inline rheometer, filling and inspection systems, lab system and coatings MES. Read-only. Predictions logged against lab, draw-down and PLC outcomes; nothing written.
-
v0.2Review console and audit log QUEUED
The console where formulators, color-matchers and quality managers approve, correct and annotate; the immutable, assurance-grade quality and VOC audit log from day one.
-
v0.3Assist mode QUEUED
Recommendations for dispersion, let-down, colorant, filtration or fill moves with the evidence and citation behind each one; approvals and corrections become labeled training events.
-
v0.4Bounded autonomy QUEUED
Guarded writes to low-risk dispersion, tinting and inspection loops inside documented formula, tolerance, VOC, equipment and rollback limits, after the quality and ROI gates.
-
v0.5Coatwin: hit color before the run QUEUED
One formula family simulated on one line and calibrated against as-made batches; pre-run path suggestions to one execution agent; a validated synthetic rare-fault set.
-
v0.6Second module in the plant QUEUED
An adjacent agent on the same plant with deeper integrations; SSO and role-based access; SOC 2 Type II in progress.
Policy
How a release reaches a line
Golden datasets and model-as-judge evaluations gate model and prompt changes in CI.
Nothing reaches a plant until its predictions have been compared to that plant's lab, draw-down and PLC outcomes in shadow mode.
Rollback by model version; a Tinteon outage returns the plant to its existing control.
Run timeline
The reference scenario each release is tested against
Every Tinteon run is an ordered, inspectable sequence. This is run_scn_01, an illustrative scenario: Plant run · one architectural deep tint base from dosing to pallet · bead mill, let-down tank, tint station, filling line. Scenario complete · batch inside spec · one human approval at the tint station.
run_scn_01
scenario · not measured
SUCCEEDED
-
01
order.receive
SUCCEEDED
0.8 s
Plant Orchestrator · Pulled the formula, the customer color standard and its tolerance, the viscosity and solids spec, the raw-material lot records and the product's VOC limit from the coatings MES and the lab system; locked the target envelope for the run.
-
02
twin.simulate
SUCCEEDED
41 s
Plant Orchestrator · Simulated 24 candidate dispersion, let-down and tint paths on the Coatwin as-made twin; ranked them on predicted Delta-E, fineness margin, viscosity, solvent use, energy and takt, and discarded every candidate that moved outside the customer tolerance or the VOC limit.
-
03
disperse.mill
SUCCEEDED
1 h 48 m
Dose-and-Disperse · Held mill energy, media temperature and residence to the predicted fineness curve from mill power and spectral reflectance, trimming inside the formula window instead of grinding to a fixed time and checking the Hegman gauge afterward.
-
04
letdown.thin
SUCCEEDED
26 m
Letdown-and-Thin · Predicted viscosity from inline rheology, temperature and solids during let-down and converged on spec with two resin additions and one solvent addition instead of the standing four, inside the formula window; no stability risk was flagged.
-
05
tint.match
SUCCEEDED
14 m
Tint-and-Color · Read the batch's spectral curve against the customer standard, predicted the colorant additions and dosed the first pass; the draw-down read inside tolerance on Delta-E after one addition, with one retint still available under the declared limit.
-
06
inspect.spectral
SUCCEEDED
continuous
Defect-and-Inspect · 12 camera and spectral streams at the plant edge: color, gloss, hiding power, fineness and seed checks on the batch, fill-weight and label checks on every pail; two seeds were attributed to one filter cartridge and Filter-and-Fill scheduled the change.
-
07
approve.human
APPROVAL
2 m 10 s
Plant Orchestrator · The lowest-solvent candidate wanted a second colorant addition beyond the declared retint limit to tighten Delta-E further. Held for the shift color-matcher, who kept the customer tolerance untouched and approved one bounded addition with a tighter draw-down check.
-
08
filter.fill
SUCCEEDED
continuous
Filter-and-Fill · Filtered through the fresh cartridge and filled pails to exact weight using fill-weight and foam feedback from inspection; the filler corrected upstream instead of rejecting downstream.
-
09
handle.palletize
SUCCEEDED
continuous
Robot-and-Handling · Moved filled pails to the pallet inside the declared safety zone with a simulation-validated handling policy; one pallet-pattern change was executed after the zone check cleared.
-
10
release.genealogy
SUCCEEDED
58 s
Quality-and-Conformance · Batch released: raw-material lots, grind profile, let-down moves, colorant additions and spectral curves, inspection outcomes, VOC and solvent use and every human decision written to the batch genealogy and checked against spec.
Tool calls
What the tool log will show per release
Every tool invocation, argument and result is written to an immutable, human-readable log, and every reasoning step is expandable. Nothing about a run is hidden from the plant. The log below is the scenario run, not a measured one.
- 05:40:02mes.read_order({ batch: "B-4471", product: "TB-DEEP-07", standard: "STD-2210" })
- 05:40:02ok formula locked · tolerance Delta-E v4 · VOC limit on product
- 05:40:03lab.read_results({ lot: "PIG-8831", test: "incoming pigment" })
- 05:40:03ok lot inside spec · accepted for dosing
- 05:40:04twin.simulate({ candidates: 24, horizon: "batch" })
- 05:40:45ok best path #9 · tolerance held · lowest solvent and energy per batch
- 05:40:46policy.evaluate({ plan: 9, stage: "bounded" })
- 05:40:46ok 7 writes permitted · 1 write requires human approval
- 05:40:47plc.write_setpoint({ tag: "MILL2.ENERGY", basis: "power+spectral" })
- 07:28:30ok Hegman on target · color strength on target · no re-mill
- 07:29:00viscosity.predict({ tank: "LD-3", basis: "inline rheology+solids" })
- 07:29:00ok two resin and one solvent addition predicted · stability clear
- 07:29:01plc.write_setpoint({ tag: "LD3.SOLVENT_DOSE", window: "formula" })
- 07:55:10ok viscosity on spec · no overshoot
- 07:56:00spectro.read_curve({ station: "TINT-2", standard: "STD-2210" })
- 07:56:01dispenser.write({ colorant: "B", basis: "spectral prediction", retint_limit: 1 })
- 08:10:12ok draw-down inside Delta-E tolerance · one retint remaining
- 08:10:13vision.stream({ cameras: 12, models: ["color_gloss", "hiding_power", "seed_foam", "fill_label"] })
- 08:21:48flag: seed · pail 0417 · attributed to filter cartridge F-2 loading
- 08:21:49defect.attribute({ event: "seed", cause: "filter.cartridge_F2" })
- 08:21:49ok filter_fill notified · cartridge change scheduled · fill held 40 s
- 08:21:52approval.request({ action: "retint.beyond_limit", reason: "tighter Delta-E on lowest-solvent path" })
- 08:24:02ok approved by the shift color-matcher · tolerance untouched · one bounded addition
- 08:24:03plc.write_setpoint({ tag: "FIL1.TARGET_WEIGHT", basis: "fill-weight+foam feedback" })
- 09:48:40ok pails on weight · no spills · labels verified
- 09:49:38mes.release_batch({ batch: "B-4471", genealogy: "full" })
- 09:50:36ok batch released · scenario complete
Reasoning trace
Goal, plan, thought, action, observation, exactly as the orchestrator would record it.
GoalDisperse batch B-4471 to Hegman fineness, let it down to visco…
Disperse batch B-4471 to Hegman fineness, let it down to viscosity spec, tint it inside Delta-E of standard STD-2210, filter and fill it to exact weight and release it with full genealogy.
PlanSimulate the whole batch on the twin first. Lead with fineness…
Simulate the whole batch on the twin first. Lead with fineness so the let-down gets the right mill base and the tint station gets the right color strength. Keep the customer tolerance and the VOC limit as hard constraints. Let the dispenser converge only inside the declared retint limit; hold anything beyond it for a human.
ThoughtThe standing practice grinds to a fixed time, lets down by fou…
The standing practice grinds to a fixed time, lets down by four additions and tints by draw-down loops. With live fineness and viscosity prediction and a spectral color-match model the batch can land on the first pass. The saving is rework, colorant and solvent, not tolerance: the customer standard is never a candidate for optimization.
Actiontwin.simulate(candidates=24) → path #9: Hegman on target in on…
twin.simulate(candidates=24) → path #9: Hegman on target in one grind, three let-down additions, one colorant pass, lowest predicted solvent and energy per batch, takt met.
ObservationPath #9 wants a second colorant addition beyond the declared r…
Path #9 wants a second colorant addition beyond the declared retint limit to tighten Delta-E further. The policy engine routes that single move to the shift color-matcher and permits the other seven writes.
ActionExecute path #9 at the bounded stage: mill energy, let-down do…
Execute path #9 at the bounded stage: mill energy, let-down dosing and the first colorant pass proceed; inspection runs continuously; the extra addition waits.
ObservationApproval granted at 08:24:02 with the tolerance untouched and …
Approval granted at 08:24:02 with the tolerance untouched and a tighter draw-down check. Two seeds were attributed to filter cartridge F-2 and Filter-and-Fill scheduled the change with a 40 second hold.
OutcomeScenario complete. The batch is released with raw-material lot…
Scenario complete. The batch is released with raw-material lots, grind profile, let-down moves, colorant additions, spectral curves, inspection outcomes, VOC and solvent use and the human decision in the genealogy. No measured plant result is claimed; this is the run the platform is designed to execute.
Roadmap
What comes after the first release
From the roadmap, by horizon.
| Horizon | Planned |
|---|---|
| 0–6 months | Three to five design partners on the wedge workflow; MVP: connectors, agent loop, review console, assurance-grade quality and VOC audit log; accuracy and re-tint, re-mill and waste ROI proven in shadow then assist; SOC 2 Type I in progress; IP-safe deployment |
| 6–12 months | Design partners convert to paid; graduated autonomy on low-risk dispersion, tinting and inspection loops; Coatwin hit-color-before-run; second module per plant; SOC 2 Type II; SSO and RBAC; first enterprise logos; quality and VOC assurance packages |
| 1–3 years | The full dose, disperse, let-down, tint, fill and inspect loop; single plant to multi-site maker to global coatings group; equipment-OEM and color-instrument channels; international expansion; benchmark data products |
| 3–10 years | The autonomous system of action for paints and coatings manufacturing, with adjacent scopes through QC and packaging across architectural, automotive, industrial and powder coatings, and a platform and API ecosystem |
Autonomy
Four stages, set per loop
A plant does not go from manual to bounded control in one step. Tinteon makes the stage explicit per loop, auditable and reversible at any time, and some decisions never leave the person.
| Stage | What the agent does | What the person does | When |
|---|---|---|---|
| Shadow | Predicts fineness, viscosity, Delta-E, hiding power, contamination and fill faults; writes nothing | Runs the plant as today; compares predictions to the lab, the spectrophotometer and PLC outcomes | First phase with every design partner |
| Assist | Recommends bounded mill, let-down, colorant, filtration and fill moves with the evidence | A formulator, color-matcher or quality manager approves or rejects each one | After accuracy is proven against the baseline |
| Bounded | Writes low-risk dispersion, tinting and inspection setpoints inside documented formula, tolerance, VOC, equipment and rollback limits | Approves anything above the limits; reviews the shift record | After the quality and ROI gates agreed up front |
| Exceptions | Escalates anything outside its envelope, any move against a customer standard or VOC limit, and release decisions | Always decides these; the agent never does | Permanent |
Accelerated computing
Physical AI, at the plant edge
A tint station reads a draw-down in seconds and a filling line moves hundreds of pails an hour, so perception has to be local and fast. Tinteon runs GPU inference at the plant edge and keeps training, simulation and optimization in the cloud, in a private cloud or on-prem. Items marked planned are on the roadmap, not running.
Edge perception
8 to 24 camera and sensor streams per line through DeepStream pipelines and TensorRT color, gloss, hiding-power, contamination and fill models; Holoscan synchronizes spectrophotometers, line-scan cameras, inline rheology, mill power and fill-weight streams at the plant edge. Design target: 50 to 150 ms per frame.
Model serving
Vision, spectral, time-series and coatings-knowledge models behind one plant router, with version routing by product family, colorant system and line. Design target: sub-second recommendations for operators, sub-250 ms holds for filling and inspection.
Training (planned)
Spectral color-match, dispersion and fineness, rheology and contamination models fine-tuned on batch telemetry, images and supervised color-matcher corrections; NeMo-based coatings reasoning. Design target: 2,000 to 5,000 GPU-hours per major model family.
Digital twin (planned)
The Coatwin as-made batch twin: GPU-accelerated spectral color-match prediction and dispersion and rheology simulation run before the batch, for color and grade changes and batch planning when a trial is expensive. Design target: 10,000 to 100,000 scenarios per design partner.
Synthetic data (planned)
Rare off-color, seed, contamination, foam, label, hiding-power and fill-fault cases generated with domain randomization, marked synthetic and validated against real plant data before use.
Optimization
Color-family batching, wash and changeover minimization, solvent and energy allocation and filling-line takt, recomputed on the GPU daily and intra-shift when the plant is disrupted.
Data engineering
Batch-genealogy, quality, equipment and rework telemetry ETL for feature generation, color correlation and cross-plant benchmarks on the GPU.
Robotic handling
Drum, pail, tote and pallet handling, with perception and motion policies built and validated in simulation before they move a container.
Guardrails
An agent that can move a setpoint needs a leash
Tinteon writes to production equipment in a chemical plant. Every capability is scoped, every write is policy-checked, every output is cited, and every action is written to an append-only audit log the plant owns.
- Bounded action space. Each agent can only write to an explicit tag allow-list, inside per-tag rate, magnitude and retint limits, at the autonomy stage the plant has set.
- Standards and limits are read, never written. The customer color standard and its tolerance, the VOC and solvent limits under the plant's permit and the batch release decision stay with the quality system and the people who own it.
- Policy engine before every write. Autonomy stage, interlock state, product, formula, shift and operator presence are evaluated before a setpoint or a colorant addition moves.
- Grounded and cited. Every recommendation carries a citation to the formula, color standard, spec or procedure it relied on. Retrieval is tenant-isolated and permission-aware.
- Immutable audit log. Append-only, exportable and retained on the plant's own storage, aligned to ISO 9001, VOC and REACH and customer color-standard record-keeping.
- Hard fallback. If the edge node, the network or a model stops, the plant keeps running on its existing mill, dispenser and filling-line control and interlocks.
- Tenant and IP isolation. Formulas, color standards, raw-material data and tolerance models never cross a customer boundary. On-prem deployment available.
Assurance posture
| Item | Scope | Status |
|---|---|---|
| SOC 2 Type I | Cloud control plane | QUEUED Roadmap, first 6 months |
| SOC 2 Type II | Cloud control plane | QUEUED Roadmap, 6 to 12 months |
| ISO 9001 · VOC and REACH · customer color standards | Grounding and record-keeping | RUNNING Design basis for retrieval and the audit log |
| Per-tenant isolation | Data, memory, models | RUNNING Design rule |
| On-prem deployment | Formula-sensitive producers | RUNNING Design option |
Tinteon supports your quality and VOC records. It is not a certification and it does not replace your quality system or your permit.
Integrations
It speaks plant, not cloud
Tinteon reads and writes through the systems already on the floor. No parallel historian, no new HMI, no rip-and-replace. Connectors are built against each design partner's stack first; long-tail integrations are out of scope for the first release.
PLC and controls of dispersers, bead, basket and horizontal mills, with mill power, media temperature and residence telemetry
Setpoint reads, guarded writes, interlocks untouched
Tint dispensers and colorant dosing systems
Colorant additions inside declared retint and additive limits
Spectrophotometers, draw-down cameras and color-standard libraries
Spectral curves and Delta-E in; cited matches out
Inline rheometers and viscometers, temperature and solids sensors, LIMS results
Continuous viscosity and stability state
Filtration skids, filling lines, fill-weight scales, labelers
Telemetry in; fill-weight and hold commands inside approved windows
Multi-camera and spectral stations for color, gloss, hiding power, seeds, foam, fill and label faults
Frames, fault events, hold recommendations
Orders, formulas, color standards, product specs, batch genealogy
Targets in, release records out
Plant and asset meters, solvent and VOC accounting; Isaac-driven drum, pail and tote cells
Hard constraints for scheduling; bounded handling
FAQ
Straight answers
The questions plant directors, formulators, color-matchers and quality managers ask in the first meeting.
Only at the autonomy stage your plant has set, and only to an explicit allow-list of low-risk loops inside documented limits. Every design partner starts in shadow mode, where Tinteon predicts and writes nothing. Assist mode follows, where a formulator or color-matcher approves each recommendation. Bounded autonomy on dispersion, tinting and inspection loops comes after the quality and ROI gates agreed up front.
They are not optimization variables. The customer standard and its tolerance, the VOC and solvent limits under your permit and the batch release decision stay with your quality procedures and your people. Tinteon reads them, cites them and writes every reading to an immutable audit log attached to the batch genealogy. It is grounded in ISO 9001, VOC and REACH and customer color standards; it is not a certification and does not replace your quality system.
Tinteon sits above your existing automation as a supervisory layer. If it stops, the plant keeps running on its current mill, dispenser and filling-line control and the interlocks it already has. A Tinteon outage degrades the plant to the way it runs today, not to a stop.
The first phase is shadow mode on one line: Tinteon reads your spectrophotometer, mill and dispenser history, inline rheology, lab results and MES while you run the plant as usual, and we measure its predictions against your lab, your draw-downs and PLC outcomes. Recommendations start when accuracy is proven against that baseline, and improve as plant-specific history and color-matcher corrections accumulate.
Formulas, color standards, tolerance models and raw-material data are tenant-scoped and never used to train another customer's models. Retrieval is multi-tenant isolated and permission-aware. Cross-plant learning, where a customer opts in, is federated and IP-isolated. On-prem deployment is available for producers protecting sensitive formulas.
You are, the same as with any control strategy. That is why every write is policy-checked, bounded, logged and reversible, every output carries a citation to the formula, standard or procedure it relied on, and anything above your threshold waits for a named approver. The audit log records the request, the reasoning, the limits applied and the human decision.
As a design partner, on one line, with one hard metric: Delta-E first-time-right rate and retints, re-mills and mill energy, solvent per batch and viscosity variance, or contamination and fill rejects. We instrument the line in shadow mode, measure the baseline, move to assist mode when accuracy is proven, and expand only when the number holds.
Full multi-module autonomy across the whole plant, long-tail integrations beyond the design partners' stacks, and self-serve onboarding. The first release is one wedge workflow, its connectors, the agent loop with human checkpoints, the review console and the assurance-grade quality and VOC audit log.
Get started
Follow the release line
Design partners see each release before it reaches their line. Ask to be one.