Platform & Safety
Hosted agent platforms
Managed runtimes and ready-made agents that do multi-step work for you.
What it is
Hosted agent platforms run AI agents for you: they provide the model, the tools (browser, shell, code editor), state and monitoring, and bill for the time the agent works. Some are frameworks you deploy your own agent onto, others are ready-made agents such as a software-engineering or browser agent.
How pricing works
Billed per hour of active agent runtime, per unit of work (credits or 'agent compute units') or per executed step. We normalise to one agent-hour of active work; waiting for a human or idle time is usually free or cheaper. Underlying model tokens may be included or billed separately: check.
We compare everything in one unit: USD per agent-hour (active runtime).
Typical uses
- Browser and form automation at scale
- Autonomous coding tasks and pull requests
- Long-running research or data-entry workflows
- Deploying your own LangGraph or custom agent without managing servers
What to compare
- Task success rate on your workflow
- Included tools and environments
- Human-in-the-loop controls
- Observability and replay
- Concurrency and runtime limits
Watch out for
- Agent hours are not equal to tasks: a slow agent can cost more per completed task than a pricier, faster one.
- Credit-based plans hide the real hourly rate; we convert at the standard published rate.
- Agents acting on live systems need permissions and review gates.
All 3 APIs, cheapest first
Free to browse| API | $ / agent-hr | Environments | Human-in-the-loop | Deploy your own agent | Quality | Price data | Compare |
|---|---|---|---|---|---|---|---|
| LangSmith Deployment (LangGraph) LangChain (LangSmith) | $0.50 | api, code | Yes | Yes | – | Unverified sample | |
| Amazon Nova Act Amazon Web Services | $4.75 | browser | Yes | Yes | – | Verified Oct 6, 2026 · aging | |
| Devin Cognition (Devin) | $9.00 | code, browser | Yes | No | – | Unverified sample |