Reef: continual learning for AI agents

Reef is a continual learning infrastructure. It sits between your agent's harness and the model that harness calls. It records each request, its response, and the feedback about the response. Reef then use these signal to produce a better version of the model weights or of the agent's harness.

The harnessruns the control loop, prompts, skills, tools, and config

requests and feedback answers and a new harness tree

Reefserves requests, records results, trains weights, and evolves the harness

inference and new weights

The modelruns the weights: local engine or hosted API

Reef serves the model, loads new weights into the engine, and sends a new harness tree to the agent.

Why Reef #

AI agents constantly accumulate valuable feedback, from passing tests to user thumbs-downs and rubric scores. Traditionally, turning these signals into a smarter agent requires a cumbersome offline pipeline: exporting logs, building datasets, retraining, evaluating, and redeploying.

Reef eliminates this friction. By integrating continual learning directly into the serving lifecycle, Reef drastically reduces the complexity of maintaining offline pipelines.

Serveforward the request and keep the exchange

Recordstore the release that produced the response

Learnthe recipe uses records to create a candidate version

Publishan accepted candidate becomes the current version

the next request is served by the new version

Existing inference engines and RL frameworks cover parts of that loop, but not the whole thing:

Ability	Inference engine (vLLM, SGLang, …)	RL training framework (slime, veRL, AReaL, …)	Reef
Serves live traffic	✓	✗	✓
Trains weights	✗	✓	✓
Versions what it served	✗	✗	✓
Stays live through updates	✗	✗	✓
Evolves beyond weights (skills, harness)	✗	✗	✓

What Reef evolves #

Reef is capable of evolve both the model weights and the harness tree, controlled by a recipe.

Model weights. Reef runs the training step and hot-swaps the result into the serving engine. See Evolve your model.
The harness tree. Reef proposes an edit, runs the current and proposed versions on your tasks, and keeps the winner. Adapters cover pi, opencode, Claude Code, Codex, DeepSeek Harness, Hermes Agent, Terminus 2, and Reef's own native harness, whose tools, hook listeners, and loop graph evolve too. See Evolve your harness.

Start using Reef #

Follow the inference and feedback quickstart to run Reef locally and submit your first report. Then choose a learning recipe for your workload and compute budget.