by lunarcave on 3/6/25, 11:58 AM with 0 comments
We built Inferable to tackle some common challenges when putting AI systems into production. It's an open-source platform with a managed control plane to help simplify the development of reliable AI applications.
After working on several LLM-based projects, we kept running into the same issues:
- Workflows crashing and losing state - No good way to pause execution for human approvals - Need to build the same boilerplate for human approvals - Difficulty managing the lifecycle of long-running processes - Can't get structured output out reliably from every LLM - Repeated effort building infrastructure for structured outputs
We built Inferable to address these with:
- Persistent workflow state that survives crashes and restarts - Built-in human-in-the-loop functionality for approvals - A distributed architecture that separates orchestration from execution - Standardized methods for extracting structured data from LLMs
Your code runs in your infrastructure while our control plane handles the orchestration. If you're familiar with temporal, think of it as temporal with higher-level LLM-native primitives.
Everything is open-source (MIT license), and self-hosting options are available if you prefer to run it all yourself.
Would appreciate feedback from anyone building AI applications with similar challenges!