Agentbase

Agentbase provides a serverless platform for building, deploying, and running persistent AI agents with minimal setup. Agents are deployed via API or web app and run in sandboxed, stateful environments that include tools, prompts, and compute. The platform uses a mixture-of-models architecture including custom models designed to reduce execution steps and costs, with three pricing tiers optimized for different task complexity levels.

• Serverless agents — Deploy agents with one API call and run them in persistent, isolated environments without managing infrastructure; start by calling the Run Agent API or using the TypeScript/Python SDKs.
• Persistent environment — Each agent runs on its own computer/sandbox so state and long-running workflows persist between sessions; enable persistence via the Persistence settings in the SDK or API payload.
• Custom tools & rules — Attach custom tools, system prompts, and rules to agents to extend capabilities; register tools through the API or SDK when creating an agent.
• Cost-per-step pricing — Pay per action, not per token, with three mode-based tiers (Flash, Base, Max); estimate costs using the pricing table and monitor in real-time with the platform's cost-tracking features.
• Mixture-of-models — Each mode employs intelligent model selection including GPT-5, Claude-4.5, and custom Agentbase models optimized to reduce agent steps and overall costs.
• SDKs & MCP support — Official TypeScript and Python SDKs and Model Context Protocol (MCP) integration simplify development and context management.
• Free beta resources — During beta, all resources like computers, browsers, and datastores are free; you only pay for agent steps.
• Quickstart & docs — Start in minutes with the quickstart and API reference; use the SDKs to run agents programmatically and iterate on agent configuration.