LeAgent
An open-source, self-hostable desktop AI agent that plans and self-corrects in a streaming think-act loop, designs and runs its own visual workflows, and streams generative UI such as KPI boards and slide decks into the chat.
LeAgent combines three capabilities in one desktop app: a streaming agent runtime that plans, calls tools, and self-corrects in a single think-act loop; agentic visual workflows, where the agent designs, runs, and refines its own flow-based diagrams with every tool automatically available as a typed node; and a generative UI layer that streams interactive interfaces, such as KPI boards, slide decks, and galleries, directly into the conversation.
It ships more than 100 built-in offline tools spanning documents, web research, data handling, code execution, databases, and media, plus a declarative rule engine, Agent Skills, and Model Context Protocol support, all running locally with SQLite and a single process by default. Its session orchestrator maintains a three-store memory, episodic, semantic, and procedural, with durable checkpoints so long-running turns can pause and resume.
It can run fully offline against a local Ollama or vLLM endpoint, or use externally provided model keys, keeping documents, sessions, and credentials on the machine unless a remote provider is configured.