Introduction

Meerkat is an open-source, provider-agnostic platform for building and running AI agents. Use rkat to process files or automate a task, embed agents in your application, or operate a coordinated system that works continuously. The same runtime supports each approach, with the capabilities you choose to include. Read how Meerkat differs from Claude Code and Codex: an operational agent platform spanning minimal automation and persistent, distributed agent systems.

​

What You Can Build With

• A full agent harness. Streaming, tool use, structured output, skills, hooks, context compaction, semantic memory, image generation and live voice. Configure built-in tools, attach MCP servers or supply your own tools.
• Your choice of models. Use Anthropic, OpenAI, Gemini, Azure and other cloud-hosted backends, or connect self-hosted OpenAI-compatible models. Choose models per agent, including different providers in the same team.
• Automation over time. Persist and resume sessions, respond to external events, schedule work and manage background jobs. These services are available when your application needs them.
• Agent collaboration. Delegate work, run specialist teams and exchange signed peer messages within one process or across machines. Mobs provide stable member identities, communication topology and repeatable workflows.
• Control over execution and access. Configure tool policy, credential bindings, delegated permissions and process confinement. Meta adds application access control and approval interfaces. See authorization for the contracts and setup paths.

The capability matrix describes build profiles and runtime capabilities; providers covers model-specific features.

​

Asynchronous Execution And Verified Lifecycles

Native asynchronous execution lets agents receive work, call tools and coordinate without tying their lifecycle to one blocking request. Durable runtime services manage recovery, scheduling and delivery when work needs to continue beyond a single run. Critical lifecycles use shared state-machine definitions that generate both TLA+ specifications and the Rust kernels used by the runtime. Model checking verifies their safety properties across the reachable states of the configured models. See machine authority for the machinery behind session execution, orchestration, scheduling and recovery.

​

Choose How Much You Need

Meerkat Core is the composable runtime, written in Rust. Compile-time features select optional capabilities; host configuration selects the services, tools and policies used by each application. This supports small embedded agents, browser/WASM applications and full-featured services in a data center. See composability for the available construction paths. The platform also brings together application operations, integration setup and shared knowledge: Use the components that serve your application. A command-line task or an embedded agent can use Core directly; larger applications can combine the platform’s services around it.

​

Use It From Anywhere

​

Continue

Start with the quickstart, explore the examples gallery, or follow the concepts overview. The architecture reference explains how execution, storage, providers and orchestration fit together.