One Gateway between your agents and their providers.
Agents call FleetWarrant's Gateway instead of holding a raw provider key. It's an OpenAI-compatible or Anthropic endpoint, not a new SDK or a sidecar to run. The Gateway forwards each call to your LLM provider with your Connector's credential and records it as an Event. Policy enforcement on that traffic is next on our roadmap, not built yet.
Your agents (examples)
- invoice-reconciler
- support-triage
- pr-reviewer
- release-bot
LLM connectors
- OpenAI
- Anthropic
- Bedrock
The call reaches your provider unchanged.
Point your agents at the Gateway
Swap your LLM provider's base URL and API key for FleetWarrant's: an OpenAI-compatible or Anthropic endpoint and an Agent API key, not a new SDK to install. Your agent needs a way to set both. Your provider account and models stay the same.
See every call, and stop any agent
FleetWarrant records every call your registered agents make through the Gateway: the tools the model asked to run, the session it belongs to, and what it cost in US$. From day one you can suspend an agent or revoke its key, stop them all at once, limit the models each one may call, and get alerts on budgets and unusual spikes.
day oneRules, when they ship
We're building a policy engine next: once it's live, you'll be able to write rules and flip them from watching to blocking, agent by agent, on your timeline. Not available yet. Today you stop an agent by suspending it or revoking its key, not by rule.