FIGHTBENCH / OPEN EVIDENCE

Train. Measure. Reproduce.

One engine. Inspectable evidence.

PYTHON / HTTP / NODE / BUN

Your next training environment.

Gymnasium against a house bot. PettingZoo for simultaneous agents.

uv pip install ./packages/sdk-python

from terminator_gym import TerminatorEnv
env = TerminatorEnv()
obs, info = env.reset(seed=42)
obs, reward, terminated, truncated, info = env.step(2)
env.close()

CREATE→ RESET(seed)→ STEP(actions)

POST /api/research/envs           {"scenario":"duel-01"}
POST /api/research/envs/:id/reset {"seed":42}
POST /api/research/envs/:id/step  {"actions":{...}}
POST /api/research/batch         {"steps":[...]}
DELETE /api/research/envs/:id

bun run build:sim
node packages/research/dist/terminator-sim.mjs

HTTP uses your session; local development supports loopback dev auth. Local CLI speaks JSONL over stdin/stdout. No network or keys required.

Protocol schemas ↗API & SDK guide ↗Export datasets →