An API and a way to run work
Decide whether requests finish quickly or continue in a worker. Check support for streaming responses, scheduled jobs, retries, and long-running processes.
Deployment guide
AI agent hosting, from first run to production
An agent host must support the work your agent actually performs. Use this guide to map the runtime, data, and safety requirements before comparing platforms.
Compare agent host platforms →What an agent host needs to handle
State and secrets
Keep conversation state, run history, and files in services that survive deployments. Store provider keys outside the agent's prompts and logs.
Safe execution
Agents that browse or run generated code need an isolated runtime with explicit limits on time, network access, and resources.
Visibility into failures
Capture request logs, tool calls, traces, and costs so you can diagnose a failed run without replaying it blindly.
Choose by workload
Chat interface and API: start with deployment platforms that support your web stack, streaming, and previews.
Research or workflow agent: prioritize persistent workers, schedules, queues, and durable state.
Coding or browser agent: compare sandbox limits, startup time, filesystem behavior, and isolation.
Compute-heavy agent: check workload limits, scaling, and whether GPU or batch execution is available.