DocuPipe,
inside your walls.
Run DocuPipe self-hosted in your own AWS, Azure or Google Cloud account, inside your VPC, or on the Kubernetes cluster in your datacenter. Documents, credentials and results stay on infrastructure you own, and nothing needs to reach the internet.
Two ways to install, one DocuPipe.
Pick the one that matches how your infrastructure is run. Both serve the DocuPipe web app and API from inside your network.
Your cloud account
Send us your account ID and we grant it access to the DocuPipe images. One command then builds the network, database, storage and service inside your account, and runs a real document through before it reports success. A fresh install takes about half an hour.
- Provisioning
- Included, from one command
- Endpoint
- Internal only, with no public listener
- Upgrades
- Rerun the same command
$ ./install.sh \ --target-account-id 123456789012 \ --aws-region us-east-1[install] DONE website / API endpoint (reachable from inside the VPC): http://internal-docupipe-onprem-7f3a.us-east-1.elb.amazonaws.com[install] all green. the bundled e2e invoice + webhook delivery test passed after the final deployment; the deployment is verified and live.
Your Kubernetes cluster
Mirror our images into your registry, fill in one values file, and install with Helm. Storage, CPU or GPU, and your model endpoint are all set in that file. The built-in test sends a document through OCR before your team starts using it.
- Provisioning
- None, it installs into your cluster
- Storage
- The S3-compatible store you already run
- Upgrades
- Change one image reference
$ helm install docupipe ./docupipe-onprem-0.4.8.tgz \ --namespace docupipe --values customer-values.yaml \ --wait --wait-for-jobs --timeout 30mSTATUS: deployed$ helm test docupipe --namespace docupipeTEST SUITE: docupipe-ocr-smokePhase: Succeeded
What you bring.
Everything the install needs from your side, so IT can prepare before the package arrives. The deployment guide has the exact permissions and settings.
Your cloud account
- A cloud account you own, new or existing, and its account ID. You send us the ID and we grant it access to the images.
- Administrator credentials for that account on a Linux, macOS or WSL machine with the cloud CLI, Terraform 1.5 or newer and Python 3.8 or newer.
- A region. US East and EU West work out of the box; other regions after a quick check with us.
- Model access enabled in the account. The installer checks it first and links you to the right console page if it is missing.
- A private path for your users and integrations: VPN, VPC peering, a transit gateway, private link or a reverse proxy. The endpoint is internal only.
- Optional: a SAML identity provider or an LDAP directory for sign-in, and a TLS certificate for your own hostname.
Your Kubernetes cluster
- A Kubernetes or OpenShift cluster with x86-64 workers, plus Helm 3 and kubectl or oc on an operator machine.
- Pull access to our chart and images, or a mirror of them in your own registry.
- A storage class for database volumes, and an S3-compatible object store with three buckets and a service account that can list, read, write and delete in them.
- A model service reachable from the cluster: its URL, the model alias and an API key stored in a Secret. Text completions are required; image input is optional.
- A hostname for the single Route or Ingress, with a TLS certificate if you serve HTTPS.
- Compute for the application pod: 4 CPUs and 8 GB of memory requested, up to 32 GB. An NVIDIA worker is optional; OCR runs on CPU by default.
Who runs what.
A private deployment is yours to operate. Here is the split, so nobody discovers it after the contract.
You run
- The cloud account or the cluster, and who can reach it over your network.
- The database and object storage, which live in your account or namespace, including backups and retention.
- The model endpoint on the cluster install, or model access in your account on the cloud install.
- The install command and each upgrade, on your schedule.
- Monitoring, network policy and sign-in: SAML, LDAP or workspace keys.
DocuPipe provides
- The install package: an installer with the infrastructure definitions for your cloud account, or a Helm chart with one values file for your cluster.
- Release images scanned for vulnerabilities before they ship, pinned by digest on the cluster install, plus pull access for your account or registry.
- Your contracted credits, loaded at install. Upgrades keep documents, results and the remaining balance.
- New releases you apply when you choose, by rerunning the command or changing one image reference.
- Support without standing access. You share logs or a screen when you choose.
- The same web app and API as our cloud, served from inside your network.
Cloud or private?
Most teams run on DocuPipe Cloud. Private deployment is for teams whose documents cannot leave their network. Same web app, same API, same schemas.
Not sure you need a private deployment? Start on DocuPipe Cloud. The free plan is open to everyone, and a private install runs the same product when you are ready.
Who runs DocuPipe privately.
Two organizations on their own infrastructure, and what the install looks like for each.
Disability claim files, read before the assessor opens them.
- Runs in
- The agency’s own cloud account
- Internet access
- Not required
- In use for
- Redacting personal details from case files
- Documents
- Hospital letters, specialist reports, functional assessments, handwritten forms
Chargebacks, invoices and credit reports, one schema each.
- Runs on
- Kubernetes in its own datacenter
- Internet access
- Not required
- Images
- Mirrored into its private registry
- Storage
- The S3-compatible object store it already runs
- Model calls
- An endpoint on its own network
Answers for your security review.
How the install is built, in the terms your security team will ask about.
- Network
- Private by default. The cloud-account install has no public subnets and no NAT gateway, and sits behind an internal load balancer with no public listener. The cluster install exposes a single Route or Ingress.
- Internet
- Not needed at runtime. The cluster install is designed for air-gapped networks.
- Our access
- None. DocuPipe has no access to your database, your storage or your network.
- OCR and models
- OCR runs inside the deployment. Model calls stay inside your own cloud account, or go to an endpoint you host.
- Privileges
- The cluster install runs under OpenShift’s restricted-v2 policy, binds only ports above 1024, and ships an optional default-deny network policy.
- Secrets
- Credentials stay in your cloud’s secrets manager or in Kubernetes Secrets. On the cloud-account install they are injected at start and never written to disk.
- Images
- On the cluster install, every image is referenced by digest in your values file, and the application image is scanned for vulnerabilities before release. Pull directly or mirror into your own registry.
- Sign-in
- SAML single sign-on or LDAP against your directory on the cloud-account install. The cluster install uses workspace keys.
- Your documents
- Never used to train or evaluate any model.
What installation looks like.
- 1
We grant access
Send us your cloud account ID, or get pull access to mirror our images into your own registry.
- 2
You run one install
One command in your cloud account. One Helm install and one values file on your cluster.
- 3
It checks itself
The cloud-account install runs a real document through before it reports success. On a cluster, the built-in test does the same.
- 4
Upgrades work the same way
Rerun the command, or change one image reference. Documents, results and credits carry over.
Questions we get from IT.
Is this BYOC, self-hosted or on-prem?
All three, depending on where you run it. The cloud-account install is what most vendors call bring your own cloud or in-VPC: DocuPipe runs inside an account you own. The cluster install is self-hosted in the strict sense, including bare metal in your datacenter with no internet connection.
Does DocuPipe need internet access?
No. Neither install needs the internet at runtime. The cluster install is designed for air-gapped networks: you mirror the images into your registry ahead of time and install from there.
Which clouds and platforms does it run on?
You can install into your own AWS, Azure or Google Cloud account, or onto Red Hat OpenShift or another Kubernetes cluster, including one on bare metal in your own datacenter.
Where do model calls go?
On the Kubernetes install, to the model service you set in the values file, on your own network. On the cloud-account install, to the model service inside your own cloud account, over your provider’s private endpoints. Either way, no document is sent to DocuPipe.
Do we need GPUs?
DocuPipe itself does not. The Kubernetes install runs OCR on CPU by default, or on an NVIDIA worker if you have one. The model endpoint you point it at is separate: a model service your team runs inside the cluster or elsewhere on your network, on whatever hardware that service needs.
How do we get support without giving you access?
DocuPipe has no standing access to a private deployment. When you need help, you share logs or a screen with us; nothing on your network opens to us.
Is it the same product as the cloud version?
It is the DocuPipe web app and API, served from inside your network, so schemas, extraction and webhooks work the way they do in our cloud. A few cloud features, such as the in-app assistant, are not part of every private install.
How is it priced?
Private deployment is an enterprise agreement with a minimum annual credit purchase, paid upfront. The credits are the same ones our cloud uses.
Bring DocuPipe inside.
Tell us where it needs to run. We’ll send the install package for your cloud account or your cluster.