The right API for
autonomous agents
Raw LLM APIs give you text generation. Agent frameworks give you orchestration. Computer Agents gives you the full stack: execution, persistence, streaming, and scaling — all managed.
Claude API
Anthropic
- Direct LLM access with streaming
- Tool / function calling
- Prompt caching and batching
- No execution environment
- No persistent file storage
- No multi-turn session state
- BYO infrastructure for agents
Pay-per-token (input + output)
Claude Code
Anthropic
- Full code execution with tools
- Local file system access
- MCP server support
- CLI tool, not a cloud API
- Runs on your machine only
- No REST API or SDK
- No scheduling or automation
Requires Anthropic API key
OpenAI Agents SDK
OpenAI
- Agent orchestration primitives
- Handoff between agents
- Guardrails and tracing
- No cloud execution environment
- No persistent workspaces
- BYO infrastructure and hosting
- OpenAI models only
Open source + OpenAI API costs
Computer Agents
Full managed platform
- Cloud execution in isolated containers
- Persistent environments and workspaces
- Real-time SSE streaming
- Multi-turn threads with session continuity
- Built-in skills, triggers, and scheduling
- TypeScript SDK + REST API
- Agent orchestration built in
- Pay-per-token with budget controls
Free tier available · Pro from $19/mo
Why developers choose Computer Agents
Six advantages that set us apart from raw APIs and BYO-infrastructure frameworks.
Managed Execution
No servers to provision. Agents run in isolated cloud containers with full file system, git, and shell access. You send a message, we handle the rest.
Persistent Environments
Workspaces survive across sessions. Install packages once, build context over time. No more re-uploading files or re-running setup scripts.
Real-time Streaming
SSE events for every tool call, file edit, and response chunk. Build responsive UIs that show exactly what your agent is doing in real time.
Multi-turn Threads
Session continuity is built in. Create a thread once and send follow-up messages that pick up where the agent left off — files, context, and all.
Built-in Skills
Web search, image generation, deep research with citations — available out of the box. No custom tool implementations or third-party wiring needed.
Pay-per-token
No idle infrastructure costs. You pay for tokens consumed, not servers running. Built-in budget controls prevent runaway spending.
Feature-by-feature comparison
A detailed breakdown across four platforms covering execution, agent capabilities, DX, and model support.
When to use which
Choose the Claude API if you...
- Need raw LLM completions without execution
- Already have your own infrastructure
- Want maximum control over every request
- Build chatbots without code execution needs
Choose Claude Code if you...
- Work locally on your own machine
- Need a coding assistant in the terminal
- Want interactive, hands-on pair programming
Choose OpenAI Agents SDK if you...
- Are locked into the OpenAI ecosystem
- Want a framework to self-host agents
- Have existing infrastructure for execution
Choose Computer Agents if you...
- Need agents that execute code in the cloud
- Want persistent workspaces across sessions
- Need scheduling, triggers, and automation
- Want a managed platform with zero infra
- Build products powered by autonomous agents
Frequently asked questions
How is this different from just calling the Claude API?
The Claude API gives you text generation — you send a prompt, you get a response. Computer Agents gives you a full execution environment: agents run in isolated containers with file systems, git, shell access, and persistent workspaces. Think of it as Claude API + infrastructure + state management, all in one API call.
Can I use Computer Agents if I already use the Claude API?
Absolutely. Many teams use the Claude API for simple completions and Computer Agents for tasks that need execution, persistence, or automation. They complement each other. Computer Agents uses Claude models under the hood.
Why not just use the OpenAI Agents SDK with my own servers?
You can — but you'll need to build and maintain the execution infrastructure, container management, workspace persistence, streaming, billing, and scaling yourself. Computer Agents handles all of that so you can focus on your agent logic.
What models does Computer Agents support?
Claude Opus 4.6, Sonnet 4.5, and Haiku 4.5. You can choose the model per agent — use Haiku for fast tasks, Sonnet for balanced work, and Opus for complex reasoning.
Is there a free tier?
Yes. You can start for free with no credit card required. The Pro plan ($19/mo) includes a 14-day free trial with full access to scheduling, triggers, orchestration, and all skills.
Ready to build with the
complete agent platform?
Start with the free tier. No credit card required. Deploy your first autonomous agent in minutes.