Flide
Fast inference engine for decision models
Developer ToolsOpen sourceOpen source · verified#ai-inference#decision-models#open-source
Listed by the ProductPerch team from Hacker News · not launched here yet
What is Flide?
Flide is an open source inference engine built for Jev-style decision models. It replaces standard generation pipelines with exact state forking, which allows the system to evaluate multiple options simultaneously while keeping every output deterministic. The engine reports speeds up to nine times faster than vLLM and SGLang.
It exposes a compatible API endpoint that accepts structured requests containing states and multiple choice questions. Users can run the software locally using Python and PyTorch, selecting from a wide range of supported models including Qwen, Gemma, and specialized decision variants. The tool also includes a verification command to test exact forking behavior on custom setups.
Flide features
- Exact state forking for deterministic multi-option evaluation
- Compatible with the Jev /v1/systemone API specification
- Supports batched requests with multiple questions per call
- Includes a verification tool to test forking accuracy
- Handles image inputs for supported multimodal decision models
- Works with LoRA adapters and mixture-of-experts architectures
Who uses Flide and for what
- Automatically categorizing and routing customer support tickets based on attached evidence
- Evaluating trading signals or market conditions using structured decision criteria
- Running compliance checks where each option requires strict deterministic validation
- Powering automated research assistants that compare multiple hypotheses side by side
Flide FAQ
Is Flide free?
Yes, Flide is released under the Apache-2.0 license and available through PyPI at no cost.
Is Flide open source?
Yes, the source code is publicly available on GitHub under an Apache-2.0 license.
What is Flide used for?
It serves as an inference engine for decision models, allowing users to evaluate multiple choices deterministically and quickly compared to standard text generation engines.
Does Flide work on Mac?
The website does not specify macOS compatibility. The project relies on CUDA and PyTorch, which are primarily optimized for Linux systems with NVIDIA GPUs.
How do I install Flide?
You can install it using Python package managers. Run uv pip install flide torchvision followed by flide serve --model [model_name] to start the service.
Comments (0)
Comments and upvotes open when this product launches on ProductPerch.
More Developer Tools launches
Monitor Laravel scheduled tasks and queues
Feel iOS haptic patterns on a real iPhone, copy the code
Deploy anywhere from your AI tool.
Python code quality analyzer for developers
Free product launch platform for indie makers
eBay listing data for apps and agents
