You choose where autobotAI runs and which models your agents use. Both choices can change later without rebuilding your agents.
SaaS or self-hosted
| SaaS | Self-hosted workspace | |
|---|---|---|
| Who runs it | autobotAI | You, in your AWS account, deployed with CloudFormation |
| Where data is stored | autobotAI's environment, isolated per tenant | Your AWS account, in the region you choose |
| Setup time | Minutes | Typically 30 to 45 minutes |
| Updates | Applied by autobotAI | You choose the version and upgrade from the autobotAI portal |
| Best for | Getting started quickly, smaller environments | Regulated data, strict data-residency and network rules |
Both run the same product: the same agents, integrations, approvals and audit trail. autobotAI is SOC 2 Type II and ISO 27001 certified; see the Trust center.
Inside a self-hosted workspace
A workspace is a dedicated deployment of autobotAI in your AWS account. Nothing is shared with other customers.
- Compute: AWS Lambda for workflow execution and Amazon ECS on Fargate for long-running work. Both scale automatically.
- Data: Amazon DynamoDB and Amazon DocumentDB hold workflow definitions, execution history, integration configuration and audit logs, in private subnets.
- Network: a dedicated VPC across availability zones, with only an HTTPS load balancer and an authenticated API exposed.
- Identity: Amazon Cognito for user sign-in, with role-based access inside the app.
- Encryption: at rest with AWS KMS keys, including customer-managed keys, and TLS 1.2 or later in transit.
Read the full architecture in Workspaces, then deploy your first workspace.
Models
autobotAI doesn't lock you to one model or provider. Connect the providers you already use and choose a model per agent, or per step inside a workflow.
| Provider | Typical reason to choose it | Setup |
|---|---|---|
| Amazon Bedrock | Models served inside your AWS account and region | Amazon Bedrock |
| Azure AI Foundry | OpenAI models on your Azure agreement | Connect your AI models |
| OpenAI | Direct access to OpenAI's latest models | OpenAI |
| OpenRouter | One key for hundreds of models from many providers | OpenRouter |
| Ollama | Models you host yourself, for data that must not leave your network | Ollama |
Need another provider? Contact support.
Choose by role, not by name
Match the model to the job rather than using one model for everything:
| Role | Good for | What matters |
|---|---|---|
| Fast model | Classifying alerts, routing, extracting fields | Speed and cost |
| Reasoning model | Investigations, attack-path analysis, deep agents | Accuracy over many steps |
| Writing model | Reports, approval summaries, pull request descriptions | Clear, faithful writing |
| Self-hosted model | Steps that touch data that must stay inside your network | Data boundary |
You can change a model later without rebuilding the agent or workflow that uses it.
What reaches the model
Agents send a model only what a step needs: the prompt, the relevant context and tool results. With a self-hosted workspace and a model in your own cloud account, such as Amazon Bedrock, that data stays within your cloud. See Bring your own model for how models are governed.