The governed runtime for production AI agent fleets
You bring the agents. Hexel makes them run together — under policy, with approvals, budgets, and a trace of every decision.
pip install hexelhexel fleet create "research" --plane fluidok fleet created on the fluid planeok policy: approval required over $50hexel run agent.py --fleet research12 actions evaluated · 1 held for approvalGoverned on the first run, not after an incident.
Hexel wraps the agent you already wrote. Policy, budgets, and approvals apply from the very first action — no rewrite, no new framework to learn.
Quickstart guideWorks with the agent stack you already use
Four systems, one runtime.
From agent to governed production in three steps.
Bring your agents
Point any agent at Hexel with the SDK or CLI — no rewrite. It joins a governed fleet instantly.
Set the rules
Define policies, budgets, and approval gates per fleet. Hexel enforces them before every action.
Run in production
Execute across Serverless, Fluid, or Dedicated planes with full traces, audit, and cost attribution.
Every action, evaluated before it runs.
Hexel intercepts each agent action and evaluates it against policy in single-digit milliseconds. Safe actions proceed, risky ones pause for a human, and everything is written to an immutable audit log.
ExploreFrom one task to ten thousand.
Fleets carry their own budgets, identities, and guardrails. Scale throughput across Serverless, Fluid, and Dedicated planes without rewriting a line of agent code.
ExploreRunning in production, under policy.
From the team.
Put your agents into production.
Start on the Serverless plane in minutes, or talk to us about Dedicated isolation and enterprise controls.