Multi-agent workflows
Chain agents into a workflow with explicit hand-offs, so each step has one job and one owner.
Agent orchestration
Rabbit Fleet Agents RFA
One console for running a fleet of AI agents — the workflows they follow, the tools they can reach, and the record of everything they did. Part of the intelligent services platforms we build.
What it is
A control room for agents, not a chat box.
Most agent work starts life as one script calling a model in a loop. That holds up until you need a second agent, a tool with real consequences, or an answer to “what did it actually do at 2am?”
Fleet Agents is the layer above that script. You describe the fleet — which agents run, in what order, and which tools each one is allowed to touch — and the console handles the running: queues, retries, schedules, and long jobs that outlive a browser tab.
Every run is recorded. Steps, tool calls, token spend, and the exact point where a person said yes or no.
Capabilities
What the console gives you
Six things that turn a working prototype into something you can leave running.
Scoped tool access
Give each agent only the tools its job needs, and see every call it made with the arguments it used.
Run observability
Traces, timings, and token spend per run, so a slow or expensive workflow is easy to find and fix.
Human-in-the-loop approvals
Pause a run before anything irreversible and route it to a person for a plain yes or no.
Enterprise-grade guardrails
Policy checks on inputs and outputs, with hard limits an agent cannot talk its way past.
Schedules and queues
Nightly jobs, long runs, and bursty workloads sit on a queue instead of an open browser tab.
When we run a fleet, the model behind it is chosen, not assumed. We pick it — and re-check it when the model changes — with Rabbit Evals: a suite that judges how a candidate model uses this console’s own tools, hand-offs, and approvals, rather than how it scores on a public benchmark. How Rabbit Evals works.
How it works
From script to fleet in four moves
01
Define the fleet
Name the agents, the order they run in, and the tools each one is allowed to call.
02
Wire your tools
Point agents at your own APIs and data using scoped credentials, not one shared key.
03
Run and watch
Start a run or put it on a schedule, then follow it live as each step reports in.
04
Review the record
Every run leaves a trace: steps, tool calls, spend, and who approved what.
Who it’s for
Built for the second week of an agent project
Teams past the prototype
You have an agent that works on a laptop and now needs somewhere real to live.
Back-office automation
Repetitive work that crosses several systems and still wants a person on the final step.
Regulated workflows
Anywhere “show me exactly what happened” is a requirement rather than a nice-to-have.
Part of our platforms
Use it on its own, or have us build it in
Fleet Agents is the agent ecosystem inside the platforms we design and build for clients. There are two ways to get it.
Use it on its own
Run it as your own console: name the agents, wire your APIs with scoped credentials, and the four steps above are the shape of the work.
See how it works — the four steps that make up a Fleet Agents setupHave us implement it
We bring Fleet Agents in as the agent ecosystem of a platform we build for you — new, or overhauled phase by phase. Every phase is scoped up front, and the source stays yours.
How we build platforms — a new platform built around Fleet Agents, or an existing system overhauled in phasesNot sure which one you need? Tell us what you are running today — one agent that outgrew a laptop, or a dozen workflows waiting on one — and we’ll say whether that is the product on its own or phase one of a platform. Talk to us about Fleet Agents.
Want to see it run?
Tell us the workflow you’re trying to hand over.