Flide
Open-source inference server for decision models that answer several questions about one input at once.
- AI agent framework
- Free
Pricing
Free and open source (Apache-2.0); you run it on your own GPU hardware.
Overview
Flide is an inference server for Jev-style decision models, which read one input and answer a set of questions about it, such as which action support should take and whether a reply is urgent. It forks the model's state after the shared input so each question is answered as if asked alone, and returns the probability of each option so you can set a threshold. It serves the /v1/systemone API and supports Qwen3.5 and Gemma 4 based models plus many published decision model families. The speedups over vLLM and SGLang are the project's own figures; the server listens on all interfaces by default, so pass --host 127.0.0.1 to keep it local.
Key features
- Exact state forking for multiple questions on one input
- Per-option probabilities in every answer
- /v1/systemone API
- Supports Qwen3.5, Gemma 4 and many decision model families
- flide verify to check forking on your own requests
Use cases
Show all 7Show fewer
Alternatives
Last updated 2026-11-09