Open Agent Kit - Build Agents in Minutes

LangCrew is a high-level framework that runs on LangGraph and takes its cues from CrewAI. You define agents, tasks, and crews with plain Python objects, and the framework handles the orchestration between them. It supports sequential pipelines, conditional routing, and parallel execution, so the same code scales from a two-agent demo to a real workflow.

The langcrew-tools package is the official toolbox, and every tool inherits from LangChain's BaseTool for a consistent interface. It covers browser automation, a sandboxed code interpreter, terminal commands, filesystem operations, cloud phone control, data fetching, search, and image generation. That saves you from writing the same integration glue that every agent project needs.

LangCrew builds approval steps into the framework rather than bolting them on. You can set interruption points before or after a task, pause for a person to review, and resume with their feedback. The bilingual UI components are designed to make those checkpoints usable in a real product, not just in a script.

A companion Agent-UI protocol and React component library turn an agent run into something you can watch. The front end shows how the agent plans, schedules work, executes tasks, and calls tools, step by step. That's the difference between staring at raw logs and actually seeing what the crew is doing.

Memory is split into short-term, long-term, and entity storage, with native PostgreSQL and Redis backends plus vector similarity search. Conversations stay coherent across threads, and knowledge carries over between sessions. For agents that need to remember context, this removes a lot of custom plumbing.

The framework streams at the token level and dispatches granular events as a run unfolds. You can follow agent thoughts, tool calls, and intermediate results as they happen, over WebSocket or SSE. That's what powers the live updates in the web UI, and it's also useful for debugging long runs.

The examples folder ships ready-to-run solutions for recruitment, marketing strategy, game content generation, travel planning, and job postings. Each one is a working multi-agent system you can read, run, and adapt instead of starting from an empty file. They double as a reference for how the framework is meant to be used.

You can run the whole stack on your own machine or server. A Docker Compose setup brings up the backend and the web chat front end together, and the only external dependency is a model API key of your choice. For teams that need to keep prompts and data in-house, that's a real advantage over hosted agent platforms.