Agents now hold your API keys and your wallet. Burnrate puts the limits outside the agent — hard budgets, action policies and on-chain enforcement — where the agent can't talk its way past them.
| Burnrate for Claude Code | live 5h/7d limits, cache-miss watchdog, a brake that blocks subagents and wide scans near the cap | burnrate · mcp-token-saver |
| Burnrate Gateway | OpenAI-compatible proxy: every agent gets its own key, USD budget and rate limit | private beta |
| Burnrate Vault | policy vault for AI trading agents on Solana: an AI risk oracle scores the token, the vault refuses the swap | solana-scam-check |
Proof — Burnrate Shield (program risk): our Solana/Anchor pipeline found two real vulnerabilities in live programs. The one filed to Immunefi — a claim-frequency over-mint in GMX-Solana's liquidity-provider program — reproduced as a working PoC that over-mints by 56.9%, and Immunefi confirmed it a valid finding (a duplicate, so no payout). A reproduced exploit, not a model's guess — the signal the vault trusts.
/plugin marketplace add burnrate-ai/burnrate
/plugin install burnrate@burnrate