BYOC Deployment
===============

Bring your own cloud: run all of Holofin in your cloud account, the application, your documents, and the AI models. Your documents never leave your account.

[How It Works](#how-it-works)
-----------------------------

In a BYOC deployment, Holofin deploys the whole platform into your cloud account: the API, the dashboard, the database, the object storage, and the models that read and check your documents. Every step of a workflow runs inside your account, under your IAM, network, and encryption policies.

1. Your systems upload a document to the Holofin API on your own domain. It is stored in your object storage.
2. Hologlyph reads the document on CPU. Holoscribe Max reads the tables and the hardest pages on your GPU.
3. Extraction and classification call the language and vision models you chose: Holofin's self-hosted models, or your cloud's AI service.
4. Your deployment stores the extracted data, check results, and audit logs in your database.

[Why Choose BYOC](#why-byoc)
----------------------------

* Processing stays in your account:

  Meets data residency rules that require documents to be processed, not only stored, in your own environment.
* Our own models:

  Hologlyph and Holoscribe Max are Holofin's models. With self-hosted language models, no third-party AI vendor sees your documents.
* Light hardware:

  One 24 GB GPU runs Holoscribe Max. Everything else runs on CPU.
* Your own security policies:

  Your IAM, network, and encryption policies apply to every component, models included.
* Same API:

  The endpoints in this documentation work unchanged. Only the base URL changes, to your own domain.

If storing documents in your country is enough and processing may happen elsewhere, a [hybrid deployment](https://app.holofin.ai/docs/api/v1.1/hybrid-deployment/) is faster to set up and needs no GPU.

[What Runs Where](#components)
------------------------------

| Component | Runs in |
| --- | --- |
| Holofin application: API, dashboard, workflows, dossiers | Your cloud |
| PostgreSQL and Redis | Your cloud |
| Object storage for documents and results | Your cloud |
| DNS, TLS certificates, and single sign-on | Your cloud |
| Hologlyph: text, layout, and reading order | Your cloud, CPU |
| Holoscribe Max: tables and the hardest pages | Your cloud, GPU |
| Language and vision models | Your cloud, self-hosted or your AI service |
| Image forensics | Your cloud |

[Document Parsing Models](#models)
----------------------------------

* Hologlyph, on CPU:

  Reads text, layout, and reading order on every page. It needs no GPU, so it scales with ordinary CPU nodes.
* Holoscribe Max, on GPU:

  Holofin's best model for table parsing. It reads the tables and the hardest pages, such as poor scans, on a single 24 GB GPU.

[Language and Vision Models](#language-models)
----------------------------------------------

Field extraction and document classification use language and vision models. You choose where they run:

* Self-hosted:

  Holofin runs its models in your account, on CPU and the same GPU. You need no AI service subscription.
* Your cloud's AI service:

  Your deployment calls an OpenAI-compatible endpoint in your account, such as Azure OpenAI, Vertex AI, or Amazon Bedrock. You pick the model and keep the contract with your cloud provider.

[What You Provide](#requirements)
---------------------------------

* A cloud account or project with admin access, where Holofin deploys the platform.
* A domain you control for the API and the dashboard.
* CPU nodes for the application and Hologlyph.
* One NVIDIA GPU with 24 GB of memory, such as an L4 or an A10, for Holoscribe Max.
* If you use your cloud's AI service: access to an OpenAI-compatible endpoint in your account.

[Compare Deployment Options](#compare)
--------------------------------------

|  | Cloud | [Hybrid](https://app.holofin.ai/docs/api/v1.1/hybrid-deployment/) | [BYOC](https://app.holofin.ai/docs/api/v1.1/byoc-deployment/) |
| --- | --- | --- | --- |
| Application, database, documents | Holofin cloud, EU | Your cloud | Your cloud |
| AI models | Holofin cloud, EU | Holofin cloud, `eu` or `us` region | Your cloud |
| Documents leave your account | Yes, Holofin stores and processes them | For inference only | No |
| GPU in your account | None | None | One 24 GB card |

[Request a BYOC Deployment](#contact)
-------------------------------------

BYOC deployments are set up per contract, including how updates to the application and the models are rolled out. To discuss your requirements, write to [security@holofin.ai](mailto:security@holofin.ai).

[Hybrid Deployment](https://app.holofin.ai/docs/api/v1.1/hybrid-deployment/)[Back to Documentation](https://app.holofin.ai/docs/api/v1.1/)