Obsigen · Deployment · as of 2026-10-22
Managed Cloud, self-hosted or on-premise. In every model, your data (the tenant) resides on dedicated infrastructure, and GRID processes tasks statelessly.
You can change the deployment model later (§4). To start, pick the one that is quickest to set up.
01 · OVHcloud
We deploy and maintain Obsigen on dedicated OVHcloud infrastructure. One contract: infrastructure, deployment and maintenance on our side.
infrastructure certifications
best forthe fastest start, no IT effort
Book a demo02 · your infrastructure
We install and maintain Obsigen in your environment. Hosting, access and security policies stay under your control.
infrastructure certifications
depends on your hosting provider
best fororganizations with their own IT and hosting
Book a demo03 · on request
The full stack - data, GRID and models - in your infrastructure. Inference with no connections to external model providers.
infrastructure certifications
depends on your data center
best forregulated sectors and highly sensitive data
Let's talk| Managed CloudOVHcloud | Self-hostedyour infrastructure | On-premise≥ 256 GB of memory for models | |
|---|---|---|---|
| data (tenant): files, database | OVHcloud instance - VPS, Public Cloud or dedicated server; EEA region | your server, VPS or private cloud | your data center |
| inference (GRID) | providers in the EEA · Zero Data Retention | providers in the EEA · Zero Data Retention | your hardware: ≥ 256 GB of memory for models (GPU or unified memory) |
| data sent to the model | only the task context - to a model in the EEA | only the task context - to a model in the EEA | none - the models run on your network |
| infrastructure certifications | ISO/IEC 27001, 27017, 27018; HDS (health data) on OVHcloud Public Cloud instances (§3) | depends on your hosting provider | depends on your data center |
| contract | one contract with Visera; OVHcloud as a subprocessor | contract with Visera; hosting on your side | contract with Visera for deployment and maintenance |
| models | the full open-weight model pool | the full open-weight pool | matched to your hardware at deployment |
| web search | a provider in the EEA - only the search query goes out | a provider in the EEA - only the search query goes out | optional, can be turned off |
| Forge and Runtime | yes | yes | yes |
| product photos (background removal) | on the tenant server available | on the tenant server available | on the tenant server |
| cost | subscription including infrastructure - one invoice | subscription; hosting on your side | project-based quote |
| status | available | available | on request |
“Memory for models” is the memory the model can access: GPU VRAM or unified memory. Model quantization, context size and the number of parallel turns change the actual threshold - which is why we set the hardware configuration at deployment, not from a table.
The certifications cover specific OVHcloud services - for sensitive data, we match the service to the requirements.
| certification | what it confirms | covers, for example |
|---|---|---|
| ISO/IEC 27001 | information security management system | VPS, Public Cloud (Compute, Storage), dedicated servers, Hosted Private Cloud - data centers outside the US |
| ISO/IEC 27017 | security of cloud services | as above |
| ISO/IEC 27018 | protection of personal data in the cloud | as above |
| HDS | health data hosting (French standard) | Public Cloud instances - not VPS; requires a higher OVHcloud support plan and activation |
| SecNumCloud | qualification by the French agency ANSSI | selected Hosted Private Cloud services |
Sources: ISO/IEC 27001, 27017, 27018 · services with HDS certification · SecNumCloud - as of October 7, 2026. The data center region in the EEA is agreed when ordering.
Getting started does not require buying hardware - you can decide on your own infrastructure later, based on real usage. We plan moving the tenant to another deployment model together with you.
We prepare each on-premise deployment individually: the hardware configuration, model selection and throughput are set at deployment. We publish performance results after measuring them on the specific configuration.
Demo · 45 min
We show Obsigen live, online, on a document you bring - with no confidential data. Then a quote and deployment on dedicated infrastructure.