LibreChat

Key features

• Every major provider in one workspace Chat through OpenAI, Anthropic, Google, AWS, Azure, Mistral, Groq, OpenRouter, and Vertex AI, plus Ollama and other local runtimes, switching models mid-conversation without losing context. Each user can also wire their own keys, so one deployment serves people with different provider accounts side by side.[1][2]

• Agents with real tools Build task-specific agents that call MCP servers, OpenAPI actions, functions, code execution, and file search, with durable prompts and conversation starters for repeatable workflows. The same agent machinery powers scheduled messages and integrations that act on the team's behalf rather than only answering in chat.[1][2]

• Retrieval over your documents Upload files and query them with file search and a RAG API offering hybrid, semantic, and full-text retrieval across local and remote sources. This turns the chat window into a query layer over manuals, policies, and project archives instead of a separate search tool.[1][2][3]

• Image generation built in Generate images from chat through DALL-E-3, GPT-image models, Azure, and Vertex AI, with editing, upscaling, and custom dimensions where the provider supports them. Visual artifacts render inline and stay attached to the conversation that produced them.[1][2]

• Multimodal conversation Send images to vision-capable models such as Gemini and GPT-4-class vision, talk by voice, and let the interface transcribe and respond in 37 translated languages. Voice input, speech output, live sync across devices, and PWA installation make the same workspace usable from desktop and phone.[1][2]

• Organization and team management Run multi-user workspaces with single sign-on, role-based access, audit logging, balance and quota controls, and per-user provider configuration. Usage analytics with prompt and token tracking show administrators exactly where consumption goes.[1][2]

• ChatGPT-parity administration Configure token management, usage balances, and quotas per user or organization, mirroring the controls teams expect from the product LibreChat replaces. Granular permission scoping decides which models and actions each group may use.[1][2]

• Private self-deployment Deploy the whole stack with Docker Compose backed by MongoDB and Meilisearch search, with no account required, no data sharing, and no training on operator data. Documented security coverage includes prompt-injection defenses, SSRF protections, PII-leakage controls, and regular dependency auditing.[1][2][3]

Who will love it

• Privacy-minded teams consolidating providers: one self-hosted workspace replaces several vendor accounts, and transcripts stay on infrastructure the team controls. Regulated or cautious organizations get audit logging and quota controls as part of that same deployment.[1][2]

• Operators already paying for model APIs: bring-your-own keys mean the software adds no seat price on top of the usage billing already in place. Cost tracking per user and organization keeps that consumption visible instead of buried in individual subscriptions.[1][2]

• Builders of repeatable AI workflows: agents, MCP servers, OpenAPI actions, and durable prompts reward people who want assistants shaped around their own processes. Conversation starters and shared presets turn one person's good setup into the whole team's default.[1][2]

Who should avoid it

• People who want zero operations: someone must run the Docker stack, configure provider keys and authentication, and apply updates, which is genuine systems work.[1][2][3]

• Buyers needing an enterprise contract: no published enterprise plan, SLA, or support tier was found, so procurement requiring those terms cannot complete from public pages.[1]

• Shoppers comparing per-seat prices: there is no seat price to compare; the real monthly cost is hosting plus provider API usage, which varies with consumption.[1][2]

What the reports suggest

Alternatives and compatibility