Inferable

Inferable is a powerful, open-source developer platform designed to simplify the creation of production-ready, LLM-powered applications. It provides the essential infrastructure and primitives for building reliable, distributed, and secure AI agents and workflows. By handling the complexities of state management, versioning, and observability, Inferable allows developers to focus on their core application logic using familiar programming languages and tools.

The platform is built around the concept of "durable workflows," which are long-running processes that can be paused for minutes or even days to await human input, and then seamlessly resume from where they left off. This makes it ideal for applications requiring human oversight or complex decision-making. As an open-source and self-hostable solution, Inferable offers unparalleled control over data, security, and infrastructure, allowing workflows to run on your own servers without exposing internal services to inbound connections.

How to use Inferable

Using Inferable is a developer-centric experience that integrates directly into your existing codebase without requiring you to learn a new framework. The process typically involves:

1. Installation: Install the Inferable SDK for your preferred language (currently TypeScript/Node.js and Go).
2. Initialization: Configure the SDK with your API credentials to connect to the Inferable control plane (either the managed cloud version or your self-hosted instance).
3. Workflow Definition: Define a workflow using the SDK. This involves naming the workflow and specifying an input schema.
4. Versioning: Create different versions of your workflow logic. This allows you to evolve and improve your application over time without disrupting in-progress executions. For example, `workflow.version(1).define(...)` and `workflow.version(2).define(...)`.
5. Implementing Logic: Within each version's definition, write your application logic. Use Inferable's LLM primitives like `ctx.llm.structured` to get structured JSON outputs from language models and `ctx.workflow.pause` for human-in-the-loop steps.
6. Execution: Trigger your workflows via an API call. Inferable's control plane orchestrates the execution, manages state, and provides observability through a developer console.

Core Features of Inferable

• Durable & Versioned Workflows: Create long-running workflows that are resilient to failures. Evolve your workflows with backward-compatible versioning, ensuring that ongoing processes complete using their original logic.
• Human-in-the-Loop: Design workflows that can pause to wait for human input or approval before continuing, enabling interactive and supervised automation.
• Structured Outputs: Reliably get structured JSON data from any LLM by defining an output schema, eliminating the need for complex prompt engineering and parsing.
• Managed State: Inferable automatically handles all the state management for your durable workflows, so you don't need to provision or manage your own databases for this purpose.
• On-Premise & Secure Execution: Run your workflows on your own infrastructure. The architecture uses outbound-only connections, enhancing security by not requiring you to open inbound ports.
• Open-Source & Self-Hostable: The entire platform is open-source, providing full transparency. You can self-host Inferable for complete control over your data and compute environment.
• Developer-First Experience: Integrates with existing programming primitives and offers SDKs for TypeScript and Go. It includes a developer console for end-to-end observability and debugging.

Use Cases for Inferable

Inferable is well-suited for a variety of complex, multi-step AI applications:

• Automated Customer Support: Create agents that handle initial customer queries, escalate to a human agent when necessary (pausing the workflow), and then resume with the provided resolution.
• Content Moderation & Analysis: Build a pipeline that analyzes user-generated content, flags potentially problematic items for human review, and takes action based on the reviewer's decision.
• Complex Data Processing: Develop workflows that fetch data from multiple sources (APIs, databases), use an LLM to enrich or analyze it, and then write the results back, with potential human validation steps.
• Financial Risk Assessment: Automate the process of gathering customer financial data, performing an initial risk analysis with an LLM, and then pausing for a financial analyst to approve or adjust the risk level.

Advantages of Inferable

The primary advantage of Inferable is its focus on production-grade reliability and developer control. It abstracts away the difficult parts of building stateful LLM applications, such as durability, retries, and versioning. The open-source and self-hosting options provide maximum security and data privacy, which is critical for many enterprises. Its use of familiar SDKs rather than a restrictive new framework lowers the learning curve and boosts developer productivity.

Pricing and Plans

Inferable offers a transparent, consumption-based pricing model with a generous free tier.

• Free Tier: $0/month. Includes 2 registered workflows, 1,000 workflow executions, BYO models, and a max concurrency of 2. Ideal for experimenting and small projects.
• Pay as you go: $10/month. Includes 20 registered workflows, 5,000 workflow executions (with additional executions at $1 per 1,000), BYO models, a max concurrency of 10, and priority email support. Suited for growing projects.
• Enterprise: Custom annual pricing. Offers custom deployments on AWS, GCP, or Azure, full data isolation, dedicated Slack support, and custom SLAs for maximum control and security.