Architecture
One gate. One trace. One claim.
Möbius is a small number of architectural primitives, enforced consistently across every surface that touches certified data.
Before a dataset can be consumed, it must be certified. Automated quality checks run on entry — null rates, grain uniqueness, pattern conformity, PII detection. A human data steward reviews the results and approves a semantic contract that captures the data's shape, its PII fields, its freshness SLA, and the decisions it is certified for. Only tables that pass both layers become certified.
Every query to certified data routes through the Möbius gateway. The gateway enforces token validity, confirms the target table is still certified, masks any PII fields the token is not scoped to see, and writes an immutable audit log entry for every access. Uncertified data has no gateway path — it is structurally unreachable.
One operating model. Two data paths.
Möbius runs the same certification workflow over every dataset. The path differs only by where the data already lives.
Warehouse-resident data
When the data already lives in your warehouse — Snowflake, Redshift, BigQuery, Synapse, Databricks, on-premise Postgres — Möbius connects via a warehouse connector, governs the certification workflow, and writes certified output back to a certified schema in your own warehouse. On Möbius Connect and Möbius Enclave, where Möbius runs inside your own environment, the data never leaves your infrastructure — Möbius holds the contract and the evidence, not the rows. This is the long-term control plane for analytical workloads.
Governed ingestion
When the data does not live in a warehouse yet — webhook payloads, API responses, CSV uploads, event streams — Möbius provides governed ingestion. Same semantic contracts, same certification engine, same steward workflow, same audit trail. This is the fastest path to value for operational and transactional data.
The full enterprise deployment runs both at once. Same certification engine. Same gateway. Same audit trail.
Every door. One answer.
Certified data is reached through four surfaces. Each one is a door into the same engine, and each one gets the same certified answer and the same refusals. There is no side entrance with looser rules.
API
REST for applications and integrations, GraphQL for field-level selection. Scoped by token, recorded on every call, masked to the same rules in every field of every query.
AI agents
Claude, Cursor, Copilot, any MCP-aware client. Agents see certified data through tools with declared scope. Uncertified data is not exposed to the agent at all.
Command line
For engineers and CI pipelines. The same certification state, the same scope checks, the same refusals — scripted rather than clicked.
Dashboard
For stewards and analysts working in the browser. A view onto the same engine, with no privileged path that the other three surfaces lack.
Every surface requires a scoped integration token. Every access is logged. The rules do not change based on who is asking.
The human in the loop, by design
Certification is not automatic. The steward workflow is deliberate.
- 1Data enters the quarantined state on ingest or connector registration.
- 2Möbius proposes a semantic contract — grain, field descriptions, PII map, freshness SLA, decision context. This is AI-assisted enrichment, not AI-granted authority.
- 3A human steward reviews the contract, approves or rejects with a reason.
- 4On approval, the automated quality checks run. If they pass, the table becomes certified.
- 5The certified dataset is written to a certified schema in your warehouse, or to Möbius's internal store for the governed-ingestion path.
- 6Downstream consumers — dashboards, pipelines, AI agents — can now reach the certified table through any of the governed surfaces.
AI assists. Humans certify. That distinction is the product.
This is the engine. It is still being built. Early access partners shape how it gets built.