Managed in Azure: Overview
DocAI Fabric can run in your own Azure subscription, inside your private network, operated by us. This page helps you choose how: how many environments, how much protection against failures, and which regions. The Setup Guide then lists the steps for your teams.
What Every Option Includes
- Your subscription. Every resource, every document and every log line stays in your Azure subscription, in the region you choose.
- Your network. The application has a private IP address in your network; your users reach it from your corporate network or VPN. Storage, secrets, the job queue and the AI models are reachable only through private endpoints.
- Operated by us. We deploy, update and monitor the application through a pipeline that signs in to your subscription without stored secrets. Every change appears in your Azure Activity Log.
- Support access to logs. Our support team has read-only access to the application's logs and metrics through Azure Lighthouse. This is required: it is how we diagnose and fix problems. It gives no access to your documents. See Application Logs.
- AI in your subscription. Today we deploy GPT-5.4-mini to classify and extract, GPT-5.4 for the in-app assistant, and Azure Document Intelligence to read the pages. These models change over time: when a better model becomes available, or your documents call for a different one, we recommend the change and make it once you accept it.
Choose a Layout
| Layout | Environments | Suits |
|---|---|---|
| One environment | Production | Evaluations and low-volume use, where configuration changes can be tried in production |
| Test and production | Test and production, in one region | Most customers: each release and each configuration change is tried in test first |
| Test and production with a recovery region | Test and production in your main region, plus a recovery setup for production in a second region | Organisations for which a regional outage must not stop document processing for long |
Any layout can have more environments: a development or user-acceptance (UAT) environment next to test, or several production environments, for example one per business unit or per region. All environments live in one subscription, each in its own resource group, with its own subnets and hostname. The recovery region works active-passive: one region serves your users at a time. See Disaster Recovery.
Choose a Level of Protection
Each level adds to the one before it.
| What fails | Standard | Zone-resilient | Regional recovery | Warm standby |
|---|---|---|---|---|
| A replica or worker crashes | Work resumes automatically, nothing lost | Same | Same | Same |
| An AI or OCR endpoint fails | Fails over to your other configured endpoints | Same | Same | Same |
| Deployment or restart | Brief interruption | No downtime (2+ replicas) | Same | Same |
| Accidental or malicious deletion | Recoverable for 90 days | Same | Same | Same |
| Datacentre (availability zone) loss | Not protected: service down, data at risk | Automatic, minutes; no data loss | Same | Same |
| Storage account outage | Down until Azure recovers | Down until Azure recovers | Switch to the replica copy | Switch to the replica copy |
| Region loss | Not protected | Not protected | Rebuild in your second region | Automatic failover to the standby |
| RTO / RPO for region loss | n/a | n/a | 4 h / 1 h | 1 h / 15 min |
What each level adds:
- Standard: a single availability zone; 90-day version history and soft delete for documents; Key Vault purge protection.
- Zone-resilient: zone-redundant storage, compute and Redis across three availability zones; two or more replicas; a durable queue.
- Regional recovery: a continuous copy of your data in a second Azure region you choose, inside your residency boundary; a scripted rebuild of the application there.
- Warm standby: a second environment, already running in that region, with health-checked traffic failover.
Recovery objectives are design targets, proven by an annual recovery test with measured results, available under NDA. The second region is chosen by you, inside your residency boundary; Azure's fixed region pair is not used unless it is the region you want. The second region's resources and AI quota are provisioned in your subscription. Active-active across regions is not offered: one region serves traffic at a time.
Recommended Configuration per Environment
| Test | Production | |
|---|---|---|
| Protection | Standard | Zone-resilient, plus Regional recovery or Warm standby if a region loss must be covered |
| Replicas | 1 | 2 or more |
| Storage | Locally redundant | Zone-redundant |
| Redis | Smallest size | Zone-redundant |
| AI capacity | A small share of the region's quota | The rest of the quota |
| Updates | Each release, first | After test, in your change window |
All environments in the same region draw on one AI quota: the quota belongs to the subscription and region, not to an environment. We split it between them and adjust the split when volumes change.
Zone redundancy, like the private network, is fixed when an environment is created. Choose each production environment's level before we first deploy it; changing it later means rebuilding the environment. Regional recovery and Warm standby can be added later without rebuilding.
Choose Regions
We can deploy into almost any Azure region that offers Azure Container Apps, Azure OpenAI and Document Intelligence. These regions have the broadest availability of the models we use:
| Geography | Region | Notes |
|---|---|---|
| Americas | eastus2 (East US 2), centralus (Central US), canadacentral (Canada Central) | |
| Europe | swedencentral (Sweden Central), francecentral (France Central) | |
| Asia Pacific | australiaeast (Australia East) | The only Australian region with Azure OpenAI, Document Intelligence and three availability zones. Recovery region: australiasoutheast (Australia Southeast) |
| Asia Pacific | japaneast (Japan East) | Three availability zones. Recovery region: japanwest (Japan West) |
Where AI processing happens
Your documents, extraction results and logs are stored in your region whichever option you choose, and OCR runs in your region. What varies is where the language models process each request. Whatever Azure stores at rest stays in your region's geography under every option.
| Option | Where requests are processed | Models and capacity | Cost |
|---|---|---|---|
| Global (recommended) | Any Azure region worldwide | The most models, new ones first; the highest quota | Pay per use |
| Data zone | Within your data zone: the United States, the European Union, or Asia Pacific | Fewer models; lower quota than Global | Pay per use |
| Provisioned | Inside your region's geography only | Fixed capacity you reserve in advance; scaling up means reserving more | A monthly charge for the reserved capacity, used or not. Costly unless the capacity is kept busy |
We recommend Global unless a rule says otherwise. Data protection rules often govern where data is stored, which Global meets, rather than where it is processed. Choose Data zone if processing must stay within your data zone, and Provisioned only if it must stay inside your country.
In Australia, the Data zone option processes requests within Asia Pacific, not only in Australia. Australia East is the only Australian region with Azure OpenAI, so Provisioned keeps processing in Australia, but an Australia East outage then stops processing until it recovers, whatever the level of protection.
What Your Teams Prepare
Every environment needs the same things from your network team: two subnets, outbound rules, DNS records, and a hostname with its certificate. A recovery region needs the same again in the second region. The Setup Guide lists them, and Disaster Recovery adds what the second region needs.
Next Steps
- Architecture and Azure Services: every Azure service we create, with its SKU, to estimate the Azure cost
- Setup Guide: the steps for your teams, and what to share with us
- Running Your Deployment: updates, the settings you manage, security
- Application Logs: what our support access includes
- Disaster Recovery: adding a recovery region, now or later; failover and recovery tests