Facts
What is Auvryn?
Auvryn is provider-neutral execution infrastructure for AI models. It validates ONNX models, runs them as durable ExecutionJobs on infrastructure it does not own, and returns verified results with provenance and telemetry.
What Auvryn does today
- ONNX model validationToday
- Auvryn accepts ONNX models, verifies their SHA-256 and checks them with the official ONNX checker before anything runs.
- Sandbox testToday
- An optional test run in an isolated container: no network, read-only root, unprivileged user.
- Durable ExecutionJobsToday
- Every run is an ExecutionJob with one lifecycle: validated, submitted once with an idempotency key, followed until it is final, and recovered after a worker restart.
- Sample Environment (simulated)Today
- A guided sample workload on a simulated environment, labelled as simulated. It never runs the model on real hardware.
- Real execution on external edge hardwareToday
- Auvryn has run ExecutionJobs on real external edge hardware through a provider integration. It is under internal evaluation and not yet offered to customers.
- Real execution on external cloud infrastructureToday
- Auvryn has run ExecutionJobs on real external cloud infrastructure through a provider integration, verified end to end in production. It is enabled for an internal workspace only and not yet offered to customers.
- One workload on more than one real environmentToday
- Auvryn has run the same reference workload on real external cloud infrastructure and on real external edge hardware through the same execution contract, with each result checked against the expected output and the evidence kept side by side. Internally verified; not yet customer-available.
- Verified resultsToday
- Result artifacts are copied into storage Auvryn controls, with their size and SHA-256. For reference workloads, Auvryn checks the output against the expected result itself.
- Provenance and telemetryToday
- Each execution records where and on what it ran, as far as it is known, and its timings and data volumes; every fact keeps its source (reported by the provider, measured by Auvryn or derived).
- Web app and Developer APIToday
- A web app and a REST API described by a public OpenAPI 3.1 document, with scoped, expiring API keys. A TypeScript SDK (@auvryn/sdk), a CLI (@auvryn/cli) and an MCP server (@auvryn/mcp) are published on npm under Apache-2.0, built on the same API; they are 0.x.
What comes next
- More providers behind the same contractNext
- Next, more cloud, edge and orbital providers behind the same provider contract as their access opens.
- Orbital executionNext
- Orbital compute is the direction Auvryn is designed for. No orbital execution has happened through Auvryn yet.
Where it is going
- Execution planningVision
- In the future, execution planning could weigh cost, latency, availability, hardware and result quality. Today, you choose the infrastructure.
What Auvryn is not
- Auvryn is not a cloud provider: it owns no data centers, satellites or ground stations.
- Auvryn does not choose the provider for you: you select the infrastructure for each run.
- Auvryn does not run workloads in orbit yet.
- Auvryn does not train models or supply data: you bring the model and its inputs.
- There is no public pricing and no paid service level agreement yet.
Short answers
What is Auvryn?
Auvryn is provider-neutral execution infrastructure for AI models. It validates ONNX models, runs them as durable ExecutionJobs on infrastructure it does not own, and returns verified results with provenance and telemetry.
Who builds Auvryn?
Auvryn is built by its founder, Jairo Rifran, in Montevideo, Uruguay.
How mature is Auvryn?
Auvryn is in early production and pre-commercial. Its API, workers and developer tools run in production, and its Sample Environment is public and simulated. Real cloud execution, real edge execution and multi-provider execution are internally verified and not yet customer-available. Orbital execution is not yet verified, and there are no paid customer deployments.
What does Auvryn do today?
Today Auvryn validates ONNX models, runs them as durable ExecutionJobs, stores verified results and records provenance and telemetry for every execution, through a web app and a REST API.
What is simulated?
The Sample Environment, the guided sample workload, is simulated and labelled as such: it never runs the model on real hardware.
What is real?
Model validation, sandbox tests, ExecutionJobs, result storage, provenance and telemetry are real. Auvryn has also run ExecutionJobs on real external edge hardware, under internal evaluation, and on real external cloud infrastructure in production, for an internal workspace only. Neither is offered to customers yet.
Can Auvryn execute ONNX models?
Yes. ONNX is the model format Auvryn accepts today; every model is checked before it can run.
Does Auvryn support edge execution?
Auvryn has executed ONNX models on real external edge hardware under internal evaluation. Edge execution is not yet offered to customers.
Does Auvryn support cloud execution?
Auvryn has executed ONNX models on real external cloud infrastructure in production, with the results verified by Auvryn, for an internal workspace only. Cloud execution is not yet offered to customers.
Does Auvryn execute AI workloads in orbit?
Not yet. Auvryn is designed for orbital compute and its provider contract can describe it, but no orbital execution has happened through Auvryn.
What is an ExecutionJob?
An ExecutionJob is one run of a validated model on a chosen provider: a durable record with one lifecycle (created, validating, ready, scheduled, running, processing, then completed, failed or cancelled), an idempotent submission, a verified result and its provenance.
What provenance does Auvryn return?
For each execution: the provider, the execution environment (cloud, edge, orbit or simulated), whether it was simulated, and the hardware, runtime and region when the provider reports them, each with its source. Auvryn never fills in what is unknown.
Which interfaces exist?
A web app and a REST API with a public OpenAPI 3.1 description, plus three 0.x packages on npm built on that API: a TypeScript SDK (@auvryn/sdk), a CLI (@auvryn/cli) and an MCP server for AI agents (@auvryn/mcp). Their source is public at github.com/auvryn/developer-tools.
Is Auvryn a cloud provider?
No. Auvryn owns no infrastructure. It is the execution layer between your model and the providers that run it.
Related
Facts reviewed 2026-10-08.