About SIMWARS
Bad tokenomics kills more crypto projects than bad code. SimWars is an attempt to change that: a swarm of language-model agents that role-play the worst actors in a token economy and stress-test your design before users do.
What's happening under the hood
Every simulation runs a roster of adversaries against your tokenomics, tick by tick. Whales accumulate and dump. Governance attackers buy voting power and submit self-serving proposals. MEV bots front-run. Sybil rings coordinate. Each agent gets the live state every tick (price, supply, staking ratio, open proposals, balances) and decides what to do on its own.
What you see at the end is a resilience report: where the design held, where it broke, which agent triggered it, and what you could change before launch.
Bring your own key
Agents call language models through OpenRouter. We don't run a shared key and we don't bill you. Paste your own key into the launcher and it's held in browser memory, or in localStorageif you tick “remember on this device.”
From there it travels over HTTPS to the API and is handed directly to the worker process that runs your sim, via an environment variable. It isn't written to any log line or stored alongside the run. The demo replays on this site never hit a model at all, they're pre-recorded event streams.
What we don't do with your key
- It's never written to disk or any log line.
- It's scoped to one worker process and dies with it when the run ends.
- The site is served over HTTPS in production.
- The code is open source. If you want to verify any of this, it's all at github.com/narasim-teja/sim-wars.