Reef
Reef is an open-source infrastructure for agents that keep improving from user feedback, for developers and researchers running model training or agent refinement.
What it is good for
- Training a model from user feedback on its answers
- Letting an agent's prompts and skills improve over time
- Serving live agent traffic while updates are applied
- Tracking versions of an improved agent before release
At a glance
- Who it is for
- Setup
- Requires programming
Level 5 of 5. You need to write code to use it.
- Cost
- Free and open source
- Type
- Framework or SDK
- Tasks
- Works with
- Slime, SGLang, vLLM, veRL, AReaL, Ollama
Momentum
- Rank this week
- Not ranked (last #42, week 43, 2026)
- Best rank
- #12 week 40, 2026
- GitHub stars
- 7,788
GitHub numbers as of Nov 2, 2026. Updated weekly.
Ranking history
Official links
Setup steps are on the official pages. We do not repeat them here, since they change.
Questions
- Is Reef free?
- Yes. It is free and open source (Apache-2.0 license).
- Do I need to know how to code to use Reef?
- Yes. You need to write code to use it.
- What does Reef work with?
- According to its own documentation, it works with Slime, SGLang, vLLM, veRL, AReaL, and Ollama.
- What is Reef good for?
- Training a model from user feedback on its answers, letting an agent's prompts and skills improve over time, serving live agent traffic while updates are applied, and tracking versions of an improved agent before release.
Similar agents
- browser-use
Browser Use is an open-source Python library and cloud service that lets AI agents control a web browser to finish tasks for you.
- AutoGen
AutoGen is a framework that lets developers build multi-agent AI applications for their own domain. It is archived and no longer maintained by Microsoft.
- CCXT
CCXT is a code library that lets developers connect to over 100 crypto exchanges and prediction markets through one interface.
- Haystack
Haystack is a Python framework that helps developers build production LLM apps and agent workflows for search and question answering.
- OpenSandbox
OpenSandbox is a sandbox runtime that lets developers run AI agents in isolated environments on their own infrastructure.
- Open Ontologies
Open Ontologies is an agent app that plans changes to a production ontology and checks their consequences for developers and data teams.