loopgym 0.1.0

OpenAI Gym equivalent for loops — create, run, benchmark, compare, evolve

LoopGym
OpenAI Gym for self-improving loops.

---

If LSS is how you declare a loop, LoopGym is how you run it.

LoopGym compiles LSS 1.0 YAML into executable environments — with deterministic simulation for CI, live model backends for production eval, and trajectory replay from LoopNet. One API. Three backends. Zero vendor lock-in on the spec.

import loopgym as lg

env = lg.make("loopbench/code-repair-v1")
obs = env.reset(task_id="cr-001")
while not env.done:
    obs, reward, done, info = env.step(agent.action(obs))

Install & run → · API reference · Quickstart script

---

Why LoopGym

Problem	LoopGym answer
Every benchmark rolls its own runner	Shared loopgym.make(env_id) registry
CI can't afford API keys	SimEnv — deterministic, free, fast
Production eval needs real models	LiveEnv — pluggable backends
Historical analysis burns budget	ReplayEnv — LoopNet trajectories, no LLM calls

LoopBench defines tasks and scores them; LoopGym executes. Clean separation, like Gym vs. benchmark suites in RL.

---

Architecture

flowchart TB
  LSS[LSS YAML spec]
  COMP[LoopGym compiler]
  SIM[SimEnv]
  LIVE[LiveEnv]
  REPLAY[ReplayEnv]
  BENCH[LoopBench runner]

  LSS --> COMP
  COMP --> SIM
  COMP --> LIVE
  COMP --> REPLAY
  SIM --> BENCH
  LIVE --> BENCH
  REPLAY --> BENCH

---

⚡ Install & run

One-liner (GitHub):

pip install git+https://github.com/KanakMalpani/LoopGym.git
python -c "import loopgym as lg; env = lg.make('loopbench/code-repair-v1'); print(env.reset(task_id='cr-001'))"

Developer setup:

git clone https://github.com/KanakMalpani/LoopGym.git && cd LoopGym
pip install -e ".[dev]"
python examples/quickstart.py
pytest tests/ -q

PyPI (after first release — PUBLISHING.md):

pip install loopgym

---

Environments

Env ID	Backend	Use case
loopbench/code-repair-v1	SimEnv	Verify-driven code repair
loopbench/research-synthesis-v1	SimEnv	Research brief synthesis
loopbench/multi-agent-debate-v1	SimEnv	Multi-agent review / debate
replay/loopnet-v1	ReplayEnv	Replay LoopNet trajectories
sim/mock-llm-v1	SimEnv	Generic mock-LLM sandbox

Bundled LSS specs live under envs/loopbench/. All validated against Loop Core Engineering in CI.

---

LoopNet replay (optional)

git clone https://github.com/KanakMalpani/loopnet.git ../loopnet
# or: export LOOPNET_SEED_PATH=/path/to/records.jsonl

env = lg.make("replay/loopnet-v1")
env.reset(record_id="ln-00042")

---

Ecosystem

Repository	Role
Loop Core Engineering	LSS / LES authority
LoopNet	Trajectory corpus
LoopGym	Runtime (this repo)
LoopBench	Benchmark orchestration

Full stack map: ECOSYSTEM.md

---

Project layout

Path	Purpose
loopgym/	Registry, envs, runtime, evaluators
envs/loopbench/	Task fixtures + LSS specs
docs/api.md	API reference
examples/quickstart.py	Onboarding smoke test

---

Citation

@software{loopgym2026,
  title={LoopGym: OpenAI Gym for LSS-Defined Agent Loops},
  author={Malpani, Kanak},
  year={2026},
  url={https://github.com/KanakMalpani/LoopGym}
}

---

_{MIT · v0.1 · Contributing · Security · PyPI}