Panicly is an AI spend firewall that sits between your application and model providers like OpenAI, OpenRouter, and Anthropic. It enforces hard budgets, token limits, model access rules, network controls, region rules, and Sentry Mode to prevent runaway traffic, bot abuse, and agent loops from creating unexpected costs. Key features include:
- Hard spend ceilings: Set workspace and project limits; stop routing when limits are reached.
- Token guard: Reject oversized requests before they hit the upstream provider.
- Sentry Mode: Hold risky traffic instantly without deploying new code.
- Model access: Approve production-safe models per project.
- Network Controls: Inspect and block network sources.
- Region Rules: Block countries and regions based on IP location.
- Audit Log: Track all configuration changes.
- Request ledger: Record every request with route, project, source, policy, decision, cost estimate, and reason.
Use cases include protecting public apps from abuse, giving client projects usage limits, stopping runaway loops, and keeping shared provider keys under policy.
