No training on enterprise workflow data
Enterprise workflow data is processed to run the workflow, not retained to train or improve models.
For teams verifying data handling, workflow control, resilience, traceability, and deployment controls before deployment. The same orchestration engine described here runs the public consumer product at momor.ai in production every day — the trust posture is not theoretical.
Trust starts with what can touch the workflow, what can be retained, and what stays inside tenant boundaries.
Enterprise workflow data is processed to run the workflow, not retained to train or improve models.
Enterprise deployments are fully isolated at the data layer — before any model sees the data. Workflow data is scoped to your tenant before it reaches providers, so documents, queries, and results stay inside the tenant boundary independently of model behavior. There is no shared cross-tenant enterprise context pool.
Provider use can be constrained by tenant policy, workflow type, or jurisdiction so workflow data only touches approved providers.
ZDR agreements cover every enterprise deployment, configured into every API call rather than negotiated per customer. That is what makes the guarantee uniform rather than a per-deployment arrangement.
Where compliance requires audit trails, logging is enabled at the tenant level — owned and controlled by the customer, stored in customer infrastructure. Query logs exist exactly where the customer configures them and flow only to systems the customer owns. Logging can also be minimized or expanded based on retention requirements.
Momor is deployed on SOC 2 Type II–audited, HIPAA-eligible infrastructure, and the vendors that process customer data hold their own independent attestations. We do not claim their certifications as ours — we publish the chain, link every trust center, and let you verify each link yourself.
| Vendor | Role | Attestations |
|---|---|---|
| Render | Hosting, PostgreSQL, Redis | SOC 2 Type II · HIPAA-eligible |
| Anthropic | AI models | SOC 2 Type II · ISO 27001 |
| OpenAI | AI models | SOC 2 Type II · ISO 27001/27017/27018/27701 |
| AI models, translation | SOC 2 · ISO 27001 | |
| Groq | AI inference | SOC 2 Type II |
| Together AI | AI inference | SOC 2 Type II |
| Moonshot AI | AI models (Kimi) | — |
| Cloudflare R2 | Uploaded files (temporary — deleted within minutes, at most 24 hours) | SOC 2 Type II · ISO 27001/27701 |
| Stripe | Payments | PCI DSS Service Provider Level 1 |
| Resend | Transactional email (primary) | SOC 2 |
| Twilio SendGrid | Transactional email (failover) | ISO 27001 |
This table is the engine's catalog, not your deployment. Search users choose their model family; enterprise tenants constrain providers by policy — a single tenant flag can restrict routing to, for example, US-based providers only. Where a vendor holds no independent attestation, the table says so plainly, and the routing constraints are how you act on it. Offering a provider is not imposing one.
Search results, weather, market data, and reference lookups are served by upstream data providers — Serper, SerpApi, Wikipedia, WolframAlpha, MedlinePlus, weather and mapping services, and similar. They receive the minimum query content needed to answer, never your account identity, documents, or workflow state. The full list lives in the Privacy Policy, and enterprise deployments can constrain or exclude any of them by routing policy.
Consumer data practices — retention schedules, data rights, location consent — live in the Privacy Policy and the plain-language Privacy Promise.
Momor is built to move the workflow forward without guessing past approval points, liability boundaries, or real judgment calls.
When the next step requires approval, clarification, or professional judgment, the workflow pauses instead of pretending automation is enough.
Missing context, conflicting records, and unresolved issues are surfaced directly instead of being flattened into a confident answer.
When a person responds, the workflow continues from the point it paused. It does not reset the chain or discard completed work.
Approval points, exceptions, and workflow branches that should stay human can be preserved as explicit boundaries in the deployment.
Momor is designed so transient failures, provider outages, and partial degradation do not collapse the workflow.
External calls retry with backoff so transient network or provider failures are absorbed before they become workflow failures.
If a provider is unavailable, rate-limited, or degraded, the system can route to the next allowed provider automatically.
If the full path is not available, the workflow can return the best useful result still supported by the remaining dependencies.
The deployment is not built around one model vendor or one external service being up at all times.
The workflow does not disappear into a black box. Activity can be reviewed after the result is delivered or the work is paused.
Which actions ran, in what order, and against which systems, documents, or sources.
Which conflicts, gaps, stale inputs, or material findings were raised during the workflow.
Which findings triggered follow-up work, altered the route, or caused the system to stop.
The chain can be inspected after the fact so teams can understand why the result finished where it did.
Need the deeper dive behind retries, failover, workflow control, and execution trees? See How It Works .
Trust also depends on what the API and SDK allow, constrain, and record during implementation.
Where compliance restricts which AI providers or jurisdictions may touch data, the routing system enforces those constraints. The customer defines the boundary; the system routes inside it.
Limit which systems, data sources, and document flows a deployment can touch.
Define where work can continue automatically and where it must return for approval.
Set what is recorded, how long it is retained, and where it is stored.
Keep required reviews, exceptions, and handoffs explicit instead of flattening them into automation.
The strongest claims below are about what does not exist: no passwords to steal, no card data to leak, no cross-tenant pool to breach.
There is no password database to breach. Sign-in is magic links that expire in 10 minutes, one-time codes, or Google and Apple sign-in. Sessions use 15-minute access tokens with rotating refresh tokens in httpOnly secure cookies.
Payments are processed end-to-end by Stripe (PCI DSS Level 1). Card numbers never touch Momor's servers, logs, or database.
Uploads are validated by file signature, not extension, with decompression-bomb protection. External pages are fetched through a proxy that blocks private-network addresses. Every surface is rate-limited.
Every model call carries layered injection defenses, with provider-specific reinforcement — an active defense built into the prompt pipeline, not a disclaimer.
TLS on every connection carrying account, workflow, document, or payment data. Encryption at rest on managed infrastructure. Auth tokens signed with HMAC-SHA256.
The same engine, gateway, and defenses run the public product at momor.ai under real traffic every day. This security posture is exercised continuously, in the open.
We run a responsible-disclosure process. We read every report and we respond. security@momor.ai — A machine-readable policy lives at /.well-known/security.txt.
We can walk through data handling, workflow control, resilience, traceability, and deployment controls in the context of your actual deployment.
Talk to us →