Manage multiple AI models, execute code, and build agents in a self-hosted platform
Imagine a browser-based interface where you switch between AI models, run code in a file-based workspace, and attach specific features as needed. LibreChat is a self-hosted AI chat platform that bundles these capabilities. It is used when organizations or individuals want to directly control their AI infrastructure on their own servers.
New in v0.8.8
The recent update, v0.8.8, adds a beta version of the agent management API, allowing agents and skills to be created or modified programmatically. It also introduces an experimental 'attached workspace' feature that enables agents to explore the file system and execute code. New features include manual summarization before the conversation context fills up and an observation tool for viewing model usage and costs at a glance. Support for GPT-6 Astra and GPT-6.1 Sol models has also been updated.
Unifying models and features
You can select and use models from major AI providers such as OpenAI, Anthropic, Google, and AWS Bedrock through a single interface. Local Ollama instances and remote APIs can also be connected, allowing you to compare various models without being tied to a specific cloud. Web search results can be reflected in conversations, and image generation and editing are included by default. You can change models or settings during a conversation or branch previous messages to continue the dialogue in a different direction.
Agents and code execution environment
A sandbox environment is provided to safely execute code, allowing you to write code in languages such as Python, Node.js, and Go and check results immediately. File uploads and downloads are handled within this environment. Through the 'agent' feature, you can create skills, which are bundles of instructions for specific tasks, or find and use community-made agents from the marketplace. It also supports connecting MCP (Model Context Protocol) servers to integrate with external tools.
Operational and security features
It supports various security methods, including OAuth2, LDAP, and email authentication, to ensure safety even with multiple simultaneous users. An admin panel allows real-time changes to user permissions and settings, enabling operational adjustments without redeployment. Scaling to multiple servers using Redis is possible, and a recovery feature ensures conversation continuity even if the connection is interrupted. System status tracking and log analysis are available through OpenTelemetry and Langfuse.
If you are ready to run your own server
This tool shines best when users have the willingness and ability to manage their own servers directly. While it can be easily installed via Docker Compose, preparing API keys for multiple AI models, checking security settings, and monitoring server resources are the user's responsibility. In particular, the 'attached workspace' and agent execution features are in an experimental stage, so they should be thoroughly tested before use with important data or in production environments. It is suitable for developers or teams who prioritize privacy and want to directly control their AI infrastructure.
