Edge-first inference
Vision, time-series and process models run on factory-edge hardware. Cloud is for training and fleet management, and is optional.
Autonomy you can put in front of an auditor
A footwear plant carries brand quality agreements, product-liability exposure and the most protected IP in the business — patterns, lasts and construction specs. Footeon is designed around that reality.
Certifications and controls
Where we are today, stated plainly, including what is still in progress.
Certifications and posture
- In progress SOC 2 Type II — audit window open [ASPIRATIONAL]
- In progress ISO 27001 — controls implemented, certification pending [ASPIRATIONAL]
- Supported GDPR — EU data residency and DPA available
- Supported Per-tenant isolation for patterns, lasts and construction IP
- Supported On-premises deployment with zero telemetry egress
Agent-specific assurances
- Graduated autonomy: observe → recommend → act-with-approval → act, set per agent and per parameter
- Human-in-the-loop approval gates on every validated bonding, lasting and moulding parameter
- Immutable, quality-grade audit log of every agent action, approval and setpoint write
- Guardrails and schema validation on tool calls; fail-closed to last known-good recipe
- Sandboxed factory-edge runtime with signed model artefacts and version rollback
- Scoped permissions per line, per station and per operator role
How your data is handled
The default is that nothing leaves the plant that does not have to.
Tenant isolation
Patterns, lasts, BOMs and telemetry are scoped per tenant and per entity. Nothing is used to train another customer’s models.
Encryption
TLS 1.3 in transit, AES-256 at rest, customer-managed keys available on enterprise agreements.
Retention
Audit and quality records retained for the product-liability window you specify; telemetry retention configurable per plant.
No-egress option
On-premises deployment with zero telemetry egress for brands whose IP cannot leave controlled infrastructure.
Deletion
Documented deletion path for tenant data and derived model artefacts on contract termination.
Every action carries its evidence
A security review usually reduces to one question: can you show me why the machine did that? Here is the answer for the one autonomous decision that paused in RUN-4417.
Action bond.window_control proposed activation 66 °C
Requested at 10:31:10, approved at 10:34:14 by the shift quality engineer, executed at 10:34:20.
Evidence Sensor readings, datasheet clause and twin prediction attached
line-b/rh-04 humidity series, adhesive datasheet AD-22 §4.2, twin-predicted bond margin at 64 °C versus 66 °C.
Record Audit trail A-88213, immutable and exportable
Written to the append-only audit store, signed, and exported nightly to the customer GRC system.
Reversibility Last known-good recipe retained
If the approval SLA had expired, the agent would have held the previous validated recipe and stopped the affected pairs rather than proceeding.
Why traces matter
Every Footeon decision that touches a cut, a mould, a last or a bond line is recorded as plan, thought, action and observation — with the material lot, sensor reading and spec clause that justified it. Quality engineers can replay any pair.
Traces are retained for the full product-liability window and exported to your MES and quality system.
How the agent runtime is bounded
Four independent mechanisms, so no single failure produces an uncontrolled write.
Before the call
- Schema and range validation on every tool call
- Parameter allow-lists per line and station
- Autonomy level checked per parameter
- Approval requirement resolved before execution
After the call
- Outcome verified by vision or metrology
- Anomalous outcome triggers fail-closed to last known-good
- Action, evidence and outcome written to the audit log
- Model regression detection against golden datasets
Who can do what
Roles map to how a plant actually works, not to a generic SaaS permission model.
| Role | View runs | Change autonomy | Approve validated params | Export audit |
|---|---|---|---|---|
| Operator | Own station | No | No | No |
| Line supervisor | Own line | Up to recommend | No | No |
| Process engineer | Plant | Up to act-with-approval | Moulding | Own plant |
| Quality engineer | Plant | No | Bonding, lasting, moulding | Own plant |
| Plant manager | Plant | All levels | No | Own plant |
| Security admin | Metadata only | No | No | All plants |
Security of the models themselves
An autonomy layer is only as safe as the artefacts it runs.
Signed artefacts
Models are signed at build time and signature-verified before the edge runtime loads them.
Evaluation gates
Golden datasets and LLM-as-judge evaluation gate every model and prompt change in CI.
One-command rollback
Any line can be reverted to a previous model version without touching the machine controls.
Sandboxed runtime
The runtime holds only the machine scopes its agents need, enforced outside the model.
Prompt-injection defence
Retrieved documents are treated as data, never as instructions; tool calls are allow-listed.
Drift monitoring
Input and output distributions monitored per line; drift raises an alert before it raises a defect.
Security numbers we hold ourselves to
Targets published so you can hold us to them.
- 0 customer pattern or last records used to train other tenants’ models
- 24h target for critical vulnerability triage
- 7yr default audit-log retention
- 100% agent outputs carrying a citation or evidence reference
Controls for regulated and brand-audited plants
Everything above, plus the identity, residency and reporting an enterprise risk committee expects.
Identity
- SAML / OIDC SSO
- SCIM provisioning
- Station-level RBAC
- Break-glass with justification
Residency
- EU, US or in-country
- Customer VPC
- Full on-prem, no egress
- Air-gapped model updates
Reporting
- Signed audit exports
- Approval SLA reporting
- Model change log
- Quarterly security review
Security questions
-
Only if you choose. Inference runs on factory-edge servers, and Enterprise deployments can run fully on-premises or in your VPC with no telemetry egress. Brand IP — patterns, lasts, construction specs — is tenant-isolated and never used to train models for another customer.
-
It stops and asks. Every agent runs at a configured autonomy level: observe, recommend, act-with-approval, or act. Bonding, lasting and any parameter under a brand or safety validation defaults to act-with-approval, and every approval is written to an immutable audit log with the sensor evidence that triggered it.
-
Agents are bounded by guardrails and schema validation, they cite the datasheet or spec clause behind a decision, and they fail closed to the last known-good recipe. Continuous evaluation against golden datasets gates every model and prompt change in CI before it reaches a plant.
-
No. Footeon is a software and factory-edge autonomy layer that sits on top of the machines you already run. We connect over OPC UA, MQTT, Modbus TCP and vendor SDKs, read sensors and vision, and write setpoints back through the controls you already trust — with approval gates on any validated parameter.
Send us your security questionnaire
We answer questionnaires, run architecture reviews with your OT team, and support on-prem proofs of concept.