File size: 2,437 Bytes
a0a20b5 5a10f7f a0a20b5 5a10f7f a0a20b5 5a10f7f 5ab1910 5a10f7f 5ab1910 5a10f7f 0292f9d 5a10f7f 1955d1e 5ab1910 5a10f7f 5ab1910 5a10f7f 5ab1910 5a10f7f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 | ---
title: Reasoning Core Environment Server
emoji: 🧠
colorFrom: green
colorTo: blue
sdk: docker
pinned: false
app_port: 8000
base_path: /web
tags:
- openenv
- agent-environment
- reasoning
- reinforcement-learning
- evaluation
- symbolic-reasoning
---
# Reasoning Core Environment
An [OpenEnv](https://github.com/huggingface/openenv) environment for formally
verifiable symbolic reasoning across logic, mathematics, planning, syntax, and
related procedural domains.
Tasks come from
[`reasoning-core/procedural-pile`](https://huggingface.co/datasets/reasoning-core/procedural-pile)
and are scored by the task-specific evaluators in
[`reasoning-core`](https://github.com/sileod/reasoning_core).
Pass `revision=` to `reset()` (or set `RC_HF_REVISION`) to pin a pile tag or commit.
## Use The Hosted Environment
```python
from reasoning_core_env import ReasoningCoreAction, ReasoningCoreEnv
with ReasoningCoreEnv(
base_url="https://huggingface.co/proxy/reasoning-core-reasoning-core-openenv.hf.space"
) as env:
result = env.reset(split="train", seed=42, size=1000)
print(result.observation.prompt)
result = env.step(ReasoningCoreAction(answer="<answer>...</answer>"))
print(result.reward)
```
Each episode has one action:
1. `reset()` returns a symbolic reasoning prompt.
2. `step(ReasoningCoreAction(answer=...))` scores the answer and ends the episode.
Plain answers and answers wrapped in `<answer>...</answer>` are accepted. Rewards
are task-specific scores in the range 0 to 1.
The environment only serves pre-generated examples from the Hugging Face
dataset. Rows whose task has no scorer in the installed `reasoning-core`
version are skipped; parked tasks that still have a scorer are served.
## Local Development
```bash
uv sync
uv run openenv validate
uv run openenv build -t reasoning-core-openenv
```
Run without Docker:
```bash
uv run server
```
The service exposes the interactive UI at `/web`, API documentation at `/docs`,
health information at `/health`, and the persistent environment API at `/ws`.
## Citation
If you use this environment, cite the Reasoning Core paper:
```bibtex
@article{sileo2026reasoning,
title={Reasoning Core: Designing Broad Procedural Data for Completion-Supervised Reasoning Training},
author={Sileo, Damien and Lacombe, Valentin and Kachler, Dimitri},
journal={arXiv preprint arXiv:2608.05148},
year={2026},
url={https://arxiv.org/abs/2608.05148}
}
```
|