FAQ
Ask the
awkward
questions.
Everything plant directors, process engineers, quality managers and security teams ask about putting an autonomous agent on a compounding vessel or a filling line — answered without hedging.
No rung is skipped. Any rung can be revoked in one action.
AUTONOMY
What the agents are allowed to do
Eventually, and only after graduated validation. Every deployment starts in shadow mode with no write access. It progresses to recommend, then approve-to-act, then bounded autonomy — per formula and per vessel, inside envelopes your process and quality owners approve.
Yes. Many sites will, at least for the first year and possibly permanently on their most sensitive products. Prediction, drift alerts and evidence packs deliver value without any write-back.
Your organisation, exactly as it is today when a technician changes a setpoint. Which is why every action carries a named approver, an envelope version and a model version in an immutable record.
Immediately, per formula and vessel, by an authorised user. The agent drops back to recommend mode and keeps predicting and learning.
SAFETY
Failure modes and limits
No. Serumon writes setpoints, never safety logic. Your safety instrumented systems, interlocks and DCS protections remain fully authoritative and untouched.
It fails to a safe state and control reverts to your existing DCS recipe within one control cycle. Losing Serumon degrades you to the status quo, never to an unsafe state.
Sensor plausibility checks and cross-sensor consistency suppress affected predictions and drop the relevant autonomy. Degraded modes are documented per sensor class in the runbooks.
Envelopes constrain not just range but rate and frequency of change, precisely to prevent that. Oscillation is treated as a first-class failure mode in design and monitoring.
GMP + QUALITY
Validation and audit
Serumon supports your GMP obligations: immutable audit, versioned models, pinned envelopes, replayable batches and documentation that can be referenced in your validation records. The certification remains yours; we are designed to be validatable inside it.
A replay: every observation, prediction, proposal, confidence, citation, envelope, approver, model version and outcome for the batch in question, exportable as a rendered evidence pack.
Model versions are immutable, signed, canaried on a single vessel class and reversible with one action. Promotion follows your change-control process with dual approval recorded.
It drafts them with enforced citations to source evidence. A human investigator verifies and signs. Where retrieval finds insufficient evidence, the system refuses rather than generating plausible text.
DATA + IP
Who owns what
It does not have to. Control runs on plant-edge GPUs with on-prem inference, per-tenant scoping and formula IP protection, up to fully air-gapped deployments where nothing leaves the site.
No. Your formulas and shade standards do not train models used for anyone else. Anonymised benchmark products require explicit, revocable consent.
You export it: telemetry, batch records, decision records and audit logs. Treatment of models trained specifically on your formulas is addressed contractually per tier.
Strict per-tenant isolation inside the deployment, with brand-scoped data, models and access. One brand’s data is not visible to, and does not train models for, another.
PRICING
Commercials
Line is $9,000 per vessel or filling line per month. Plant is $60,000 per month for all five agents plus the twin. Enterprise is custom, typically landing between $400k and $5M ACV. Annual prepay carries a 15–20% discount.
No. Pilots are paid, and fees are credited against the first subscription term on conversion. Free pilots have no owner, no urgency and no conversion path.
Plant-edge GPU hardware, connector and integration work, and any sensor retrofit. These are scoped during the site survey and quoted separately rather than hidden in the price.
Yes, on the workflows that can be measured cleanly: shade holds, giveaway, changeover and yield. Outcome components are negotiated against the baseline recorded during the pilot.
DEPLOYMENT
Getting it running
Instrumentation typically takes about four weeks, then shadow mode produces measured prediction accuracy and a real baseline within a further eight. Control benefits follow validation, which depends on your batch frequency.
OPC-UA or Modbus access to the relevant equipment, retained historian data, LIMS export for assurance workflows, an identity provider for SSO, and a security path for the edge nodes.
Yes, during survey, instrumentation and go-live. Manufacturing autonomy built entirely remotely is manufacturing autonomy that fails on the night shift.
That is the intended path: land on one line, expand to the plant, then standardise across sites under an enterprise agreement.
INTEGRATION
Fitting the existing stack
No. Those remain your systems of record. Serumon is the system of action: it perceives, decides and writes setpoints, and it integrates with both rather than displacing them.
Often, through Modbus or an edge SDK adapter. Older equipment with coarse setpoint resolution limits how much control benefit is available, and we will tell you that during the survey rather than after the contract.
Yes. The edge SDK, sandbox and adapter documentation are available to customer teams and certified integrators.
Yes. The control path never depends on the cloud, and air-gapped deployments receive signed model artefacts through your existing controlled process.
COMPANY
About Serumon itself
Not yet. Serumon is pre-revenue and recruiting three to five design partners. Every scenario and figure on this site is labelled as modelled or aspirational for exactly that reason.
You should be sceptical. The mitigations are structural: Serumon is additive rather than load-bearing, failure degrades to your existing recipe, data is exportable, and early stages require no write access at all.
Contractual continuity provisions, data export rights and the on-prem deployment option are the practical protections, and we are happy to discuss them explicitly in enterprise agreements.
Equipment, MES, LIMS, ERP, safety systems, and any vertical outside cosmetics and personal care. Focus is the strategy, not a limitation we are apologising for.
STILL DECIDING?
The four things to check
STILL HAVE QUESTIONS
Ask us something harder
If the answer is not here, it is probably a question we should be answering publicly. Send it and we will answer directly — and likely add it to this page.