Best open-source alternatives to Claude Managed Agents

Omnara is an open-source platform for managing and running AI agents. It is aimed at developers and teams that want to launch agents through an API, the CLI, a dashboard, or an SDK, while keeping control over models, tools, permissions, and execution environments. The project positions itself as a self-hostable alternative to hosted agent platforms. The system is built around a REST API, a worker-based execution model, and persistent storage for agent state. It can run agents against multiple model providers, use built-in or custom tools, and execute tasks on connected machines or sandboxes. The README also describes integrations for Slack, Discord, GitHub, and scheduled triggers, making it suitable for automated workflows and collaborative agent operations.

Centaur is a self-hosted agent platform aimed at teams that want to centralize AI agent usage instead of running many isolated local setups. It integrates with Slack as the primary chat surface, while also exposing an HTTP API for creating sessions, sending messages, starting executions, and streaming events. The system assigns each conversation its own Kubernetes-based sandbox with a shell, workspace, and access to approved tools. It is designed for durable, auditable agent work: conversations can persist state, resume after interruptions, run long-lived workflows, and reconnect without losing output. Centaur also supports organization-specific overlays for tools, workflows, personas, skills, and prompts, making it suitable for operational teams that need consistent agent behavior and controlled access to internal systems.

Letta is a project for building stateful AI agents that retain memory and improve over time. It is aimed at developers who want to run agents locally or self-host them, and it also connects to browser-based and messaging-channel experiences. The README points users to the current codebase, which includes an agent harness, terminal UI, App Server, channels, and the runtime used by the desktop and web apps. It can be installed from npm, launched as a local CLI, or run as a server for self-hosted agents. The project also integrates with Letta Cloud to keep agent memory, identity, and conversations available across computers, and it provides documentation plus an agent SDK for TypeScript applications.