StackMap
Subscribe

gym-anything vs oransim

CMU framework that turns real software — browsers, IDEs, EMRs, CAD — into standardized agent environments: start the app, hand the agent a task, score it with automatic verifiers. — versus — Open causal engine for marketing simulation: a virtual consumer society with LLM personas answers do()-style counterfactuals — rank campaign combos, swap KOLs mid-flight, replay spend. Apache-2.0.

The curated verdict

Two kinds of synthetic environment for agents: gym-anything wraps real software (browsers, IDEs, EMRs) into standardized tasks, oransim simulates a consumer society to answer do()-style counterfactuals. Environments to act in vs a world to ask questions of.

gym-anythingoransim
Stars2771.1k
Forks39133
LanguageShellPython
LicenseMITApache-2.0
Last activity7 days ago1 months ago
Topicsevals, agentsagents
Curated connections32

gym-anything — the curator's take

The missing middle layer for computer-use agents: a Core runtime (environment lifecycle, actions, observations, verifiers), a benchmark collection wrapping real applications like Moodle, and reference agents (Claude, Gemini, Qwen, Kimi) — three parts connected by contracts, each independently replaceable, one CLI to run it all. The `doctor` setup command and environment caching signal real operational care. Use it to evaluate your CUA agent beyond browser-only benchmarks, or to gym-ify internal software by adding a task folder with a setup script and a checker. NOT an agent framework — the agents are references, bring your own. Young (arXiv 2026, CMU L3): environment coverage is the current bottleneck.

oransim — the curator's take

Rare thing: a production causal-ML engine open-sourced end to end. Pearl-style SCM (64 nodes/117 edges), per-arm counterfactual heads in the TARNet/Dragonnet lineage, causal Neural Hawkes rollouts, and LLM 'soul personas' that read your actual creatives — every layer carries inline paper citations and any prediction is traceable through the graph. Know what you're holding: the audit artifact of a Shenzhen martech company's open-core play. The OSS ships a 21k-note demo corpus plus synthetic data; real predictive power needs their licensed 小红书 panel or your own DataProvider, and the research-grade models are code-complete but weights-pending. Use it to study causal agent-based simulation or as a BYO-data scaffold — don't expect turnkey marketing truth.