What it does
Inferable is a managed durable execution runtime for AI workflows with humans in the loop. Workflows run in your own infrastructure, including behind firewalls or private VPCs, and connect through long polling so no inbound ports need opening. When a workflow's input schema or logic changes, you create a new version and Inferable keeps running executions on the version they started with, so new versions can roll out gradually. Approval steps can notify people by Slack or email and preserve context, and structured outputs from an LLM are parsed, validated and retried automatically. The repository also lists memoized results, a timeline view for observability and agents that use tools. SDKs are available for Node.js and TypeScript and for Go, with a .NET SDK marked experimental. The repository holds the control plane, a playground front end and management console, and an alpha command-line tool. The project says it can be fully self-hosted, and all code in the repository is MIT licensed.
Capabilities
- Durable, versioned workflows that keep running executions on their original version
- Human-in-the-loop approvals through Slack or email
- Structured LLM outputs with automatic parsing, validation and retries
- Workflows execute in your own infrastructure using long polling
- Node.js and Go SDKs, with an experimental .NET SDK
- Timeline view for observability
When to use it
- Building AI workflows that need a person to approve risky actions
- Running long-lived workflows that must survive logic or schema changes
- Keeping execution inside a private network or VPC
- Getting typed, validated outputs from an LLM
