Your model account
Azure OpenAI or Azure AI Foundry in your own subscription and region, Bedrock in your AWS account, or Google Gemini — all behind the gateway.
- Your quota, your bill
- Your region
Private AI deployment
Private AI on Azure, AWS, Google Cloud — or on‑premises.
Any cloud, any region, disconnected sites included — built for mission-critical work at geo scale.
Botminds is one platform with one release, and it deploys wherever your policy says: our managed cloud, a dedicated environment, your own Azure, AWS or Google Cloud account, or your own data centre — down to disconnected sites that take no inbound access. It scales out on Kubernetes, runs in as many regions and geographies as you need, and keeps every document, index and record in the region you choose.
Azurein your account and region
AWSin your account and region
Google Cloudin your account and region
On-premisesin your data centre, even disconnected
Your Botminds
- Agents
- Collections
- Experiences
- Governance
Your region · your keys · your identity provider
What is private AI?
Private AI is AI that runs inside a boundary you control — your cloud account, your data centre or an environment dedicated to you — so your documents, prompts and results stay under your rules, in your region. With Botminds, private AI is a deployment choice, not a different product: the same platform runs on Azure, AWS, Google Cloud or on‑premises, in any region, with the models you approve.
Scales out on KubernetesServices and workers add capacity as work arrives, and give it back when it is done.
Built for millionsMillions of documents, records and agent runs — the indexes and workers grow with them.
Any cloud, any regionAzure, AWS, Google Cloud or on-premises, in the region your data must stay in.
Always onRedundant services, rolling releases and continuous backups, for mission-critical work.
Secure and governedSingle sign-on, encryption, least privilege and a sealed audit trail.
Deployment shapes
Private AI deployment: five shapes, one platform.
A bank, a hospital and a defence supplier draw the boundary in different places: who operates the system, whose account it runs in, what may reach it from outside. Each shape runs the same platform; what changes is where the boundary sits.
01fastest start
Our managed cloud- Runs in
- Botminds’ cloud, in the region you choose
- Kept apart by
- Isolation per customer, down to the indexes
- Operated by
- Botminds
- Updates
- Zero-touch rollout by us
02single tenant
Dedicated environment- Runs in
- A stack of its own, in the region you choose
- Kept apart by
- Its own network, key vault, database and storage
- Operated by
- Botminds
- Updates
- Zero-touch rollout by us
03Azure · AWS · Google
Your cloud account- Runs in
- Kubernetes in your own subscription
- Kept apart by
- Your account boundary and your network rules
- Operated by
- Your team, us, or both
- Updates
- Checksummed bundles of signed, verified images, on your schedule
04on-premises
Your premises- Runs in
- Kubernetes in your data centre
- Kept apart by
- Your building, your network
- Operated by
- Your team, with our support
- Updates
- Checksummed bundles of signed, verified images, on your schedule
05behind your airlock
Disconnected- Runs in
- Your cloud or premises, behind an airlock
- Kept apart by
- No inbound access and no VPN to us
- Operated by
- Your team, with our support
- Updates
- Checksummed bundles carried across your airlock; images from our signed registry
Same product in every shape. Agents, workflows, meshes, collections, experiences and governance behave the same in each. A solution built in one shape moves to another as a package, with its secrets stripped and re-entered on the other side.
Where it runs
Azure, AWS, Google Cloud or on‑premises: what runs inside.
The platform is packaged for Kubernetes — the runtime every major cloud and most data centres offer. Wherever the cluster sits, the same services start, on your bill, inside your network, signed in through your identity provider.
Kubernetes · the same services in every shape
- Web appthe screens people usescales on load
- APIevery request, every permission checkscales on load
- Agent runtimeagents, meshes and the run recordscales on load
- Workflow runnersscheduled and triggered workflows
- Workersdocument work, scaled on queue depthscales on queue
- Model gatewayevery model call; holds the provider keyscales on load
- Document conversionoffice files, with no network at all
- Search indexkeyword and vector search
Data, in your region
- Databasepoint-in-time restore; backups can go to a paired region or a second site
- File storagedocuments and page images, with versions and recoverable deletes
- Key vaultsecrets live here, never in a package or an image
At scale
Private AI at scale: Kubernetes autoscaling, from node to geography.
A private deployment is built for the same internet-scale workloads as our own cloud: millions of documents, thousands of agents and users. Every layer scales out on its own signal, and adding capacity means adding nodes.
loadServices scale on load
Web app, API, agent runtime and model gateway are stateless: copies are added as load rises and released as it falls.
queueWorkers scale on queue depth
Document work waits in a durable queue; workers are added as it grows and released when it drains.
nodesMachines scale with demand
The cluster adds nodes when new copies need room, and gives them back after.
One region · Kubernetescopies of each service spread across nodes
node
- Web app
- API
- Gateway
node
- Agents
- Workers
- Workers
node
- API
- Agents
- Gateway
node
- Web app
- Workers
- Workers
node
- Agents
- API
- Workers
new node joining
- Workers
- Workers
Durable queuework is held outside every process — a lost node’s documents resume on another
Mission-critical availability
Geo-level availability for mission-critical private AI.
Run a deployment in every geography you serve, each keeping its own data at home. Each deployment can back up to a paired region in the same geography and recover there.
Europe
Europe · primary regionthe full platform · data stays in this geography
Paired European regionrestore point, ready for recovery
North America
United States · primary regionthe full platform · data stays in this geography
Paired US regionrestore point, ready for recovery
Asia Pacific
India · primary regionthe full platform · data stays in this geography
Paired Indian regionrestore point, ready for recovery
Your site
On-premises · disconnectedthe same platform, behind your airlock
Your backup siteon your own schedule
- Backed up continuouslyPoint-in-time restore; each deployment can keep copies in a paired region.
- Region lostThe database is restored in the paired region from its backup.
- Work resumesQueued documents carry on from where they stopped.
Availability, measured where it runs. Every few minutes the platform scores its own capabilities from its own database and keeps the result as a ledger, with a page, an export and a quarterly report — nothing has to reach Botminds. Two tools ship with every bundle: one recomputes the figures with a read-only account, the other checks a report’s digest offline. You can check our number without trusting us.
Data residency
Data residency: everything stays in your region.
Loan files stay in the country; health records stay in the region. Every store the platform keeps sits in the region you choose, and nothing leaves it except what you choose to send.
Your regionbackups: a paired region, or a second site of yours
- Original documentsand every page image
- Extracted valuesfields, records and corrections
- Search indexeskeyword and vector
- Conversationsand attachments
- Run recordsevery step, every approval
- Audit trailsealed, per project
Leaves the region by your choice
Model callsto the model deployments you approve — including your own, in your region
Resultsto the systems you connect, by API and webhook
Notification emailsthe ones you choose to send, through the mail service you configure
Sovereign AI, in practice. Sovereign AI means keeping control of your data, the models that read it and the infrastructure under both. The practical test: name every place your data goes. In a private Botminds deployment that list is your region, the paired backup region you choose, the model deployments you approved, the systems you connect and the mail service you configure.
Models in your boundary
On-premise LLM choices: the models stay yours too.
In Botminds, models are records you configure, and every call goes through the model gateway inside your deployment — so you decide which models agents use and where they run.
Reached privately
In a dedicated Azure environment, your model account is reached over a private endpoint inside the deployment’s network.
Models you run
Point the gateway at a model deployment on your own hardware, and a disconnected site keeps AI on site. Agents use it like any other model.
The key lives in the gateway. The model gateway is the one service that holds the provider key; every other service reaches the model through it. How models are scoped, pooled and paid for is in the LLM gateway.
Updates
Updates for a private deployment, even a disconnected one.
Every release is built from signed, verified images. In our cloud it rolls out zero-touch; at your site it arrives as a checksummed bundle and rolls out on the schedule you set. A disconnected site receives the bundle across its airlock and pulls the images from our signed registry — checked on our side before the bundle leaves and on yours before it runs.
Our side
- Proven firstEvery release is validated on our own staging and production before it ships.
- Signatures checkedEvery image in the release must carry a valid signature. One unsigned image fails the export.
- BundledChart, database changes for exactly that release, and checksums. No secrets inside.
Your side · one upgrade command
- Verifychecksums match the bundle
- Check secretsany new setting the release needs is already in your vault
- Back upthe database, on demand, before anything changes
- Migrateexactly the database changes it is missing
- Roll outon the schedule you set, recorded in the availability record
- Proveeverything is healthy before it reports done
Network, identity and keys
Private endpoints, and one identity per service.
In a dedicated environment the data services sit on private endpoints, each service runs under an identity of its own, and every secret sits in a key vault you can audit.
01
Private endpoints
Database, file storage, cache and model account are reached over private endpoints, with public network access switched off.
02
An identity per service
Web app, API, agent runtime, workers and jobs each run under their own identity, granted exactly what that service needs.
03
Secrets in your vault
Keys and connection strings live in a key vault, loaded at start — never in a package, an image or a release bundle.
04
A sealed converter
Office files are converted in a service with no network, no secrets, no identity and a read-only file system.
In every shape. AES-256 encryption at rest, TLS on the wire, SSO and SAML, and role-based access down to the collection. Certifications — SOC 2 Type II, ISO 27001 — are on security at Botminds; how agents are kept in bounds is in AI agent security.
Side by side
Private AI vs public AI services vs building your own.
Three ways to get AI behind your own boundary: accept a public service’s terms, assemble an on-premise AI stack yourself, or deploy a platform built to run privately.
| Public AI service | Build your own stack | Botminds private AI | |
|---|---|---|---|
| Where data lives | The vendor’s cloud and regions | Wherever you put it | The region you choose, in your account or premises |
| Clouds | The vendor’s | Whatever you build for | Azure, AWS, Google Cloud or on-premises |
| Disconnected sites | No | If you build the tooling | Checksummed bundles of signed, verified images, carried across your airlock |
| Scale and availability | The vendor’s | Yours to engineer | Kubernetes autoscaling, multi-region, paired-region backups |
| Models | The vendor’s choice | Yours to wire and govern | Your accounts and deployments, through one gateway |
| Agents, workflows, apps | Fixed features | Yours to build | Built in, the same in every shape |
| Updates | When the vendor ships | Yours to engineer | Signed, verified images: zero-touch in our cloud, checksummed bundles on your schedule at your site |
Build or adopt? An on-premise LLM is the easy part. The agents, retrieval, review screens, approvals, audit and upgrades around it are most of the work. The build vs adopt whitepaper sets out the costs.
How to choose
Choosing a private AI deployment: five questions.
Answer these in order; the first “yes” usually decides the shape.
- No inbound access at all?
A disconnected site on your cloud or premises: no inbound access, no VPN to us, each release a checksummed bundle carried across your airlock.
- Hardware must be yours?
Kubernetes in your data centre, the same release as the cloud.
- Must run in your account?
Your Azure, AWS or Google Cloud subscription, on your bill and network rules.
- Must be single-tenant?
A dedicated environment with its own network, vault and data stores, run by us.
- None of these?
Our managed cloud, in the region you choose — the quickest start.
Questions people ask
Private AI deployment, answered.
More from the platform: the full architecture, and how security works at Botminds.
What is private AI?
Private AI is AI that runs inside a boundary you control — your cloud account, your data centre or a dedicated environment — so your data, prompts and results stay under your rules and in your region. With Botminds it is a deployment choice: the same platform runs on Azure, AWS, Google Cloud or on-premises.
Can Botminds be deployed on-premises?
Yes. The platform is packaged for Kubernetes and runs in your own data centre with the same release as the cloud, operated by your team with our support.
Which clouds does Botminds run on?
Azure, AWS and Google Cloud, in your own account and in any region, as well as our managed cloud and on-premises installs.
Can Botminds run in a disconnected site?
Yes. A disconnected site takes no inbound access and needs no VPN to us. Its images are pulled from our signed registry, and each release arrives as a checksummed bundle carried across your airlock. Every image is signature-checked before export, the bundle carries no secrets, and one upgrade command on your side verifies the checksums, backs up, migrates and rolls out the release.
How does a private AI deployment scale?
On Kubernetes: stateless services add copies as load rises, workers add copies as the queue grows, and the cluster adds nodes as copies need room. Copies of each service are spread across nodes, and you can run a deployment in each region or geography you serve.
Can I run an on-premise LLM with Botminds?
Yes — a model deployment on your own hardware reached through the model gateway, or Azure OpenAI, Azure AI Foundry or Bedrock in your own account and region. Agents use it like any other model. See the LLM gateway.
How does Botminds handle data residency?
Documents, page images, extracted values, search indexes, conversations, run records and the audit trail all stay in the region you choose. Each deployment can back up to a paired region in the same geography and recover there — or, on your premises, to the second site you choose. Data leaves the region as model calls to deployments you approve, results to systems you connect, and notification emails you choose to send.
What is sovereign AI?
Sovereign AI means keeping control of your data, the models that read it and the infrastructure under both — often within one country. A private Botminds deployment gives you that control: your region, your account or premises, and your choice of models.
How is a private deployment kept highly available?
Core services run as several copies spread across nodes, workers scale with the queue, and queued work survives the loss of a machine. The database has point-in-time restore, and each deployment can back up to a paired region in the same geography (or a second site of yours) and recover there.
How do updates work in a private deployment?
Every release is built from signed, verified images. In our cloud it rolls out zero-touch; at your site it arrives as a checksummed bundle, rolled out on your schedule. One upgrade command verifies the bundle, checks your vault for new secrets, backs up the database, applies the database changes it is missing and confirms the system is healthy.
Is my data used to train AI models?
No. Your documents never train shared or third-party models. See security at Botminds for certifications and controls.
Learn the basics
Data residency for AIResidency, sovereignty and localization, GDPR transfers and model calls→On-premise LLM deploymentGPU sizing, serving stacks, open-weight models and trade-offs against cloud→Sovereign AIWho controls the data, models, compute and law under an AI system→All guidesEvery explainer, by topic→
Private AI, wherever your policy says.
Tell us your boundary — cloud, region, geography, premises or airlock. We’ll show you the deployment that fits it.