Product overview
Libre WebUI brings local and cloud models into one self-hosted workspace for conversation, action, evidence, collaboration, and creation. Start with a quiet chat. Add only the capabilities your work needs.
Stream a simple answer or carry a long piece of work across models. Libre keeps the conversation legible while exposing the controls that materially change what a model sees and how it responds.
Ollama is the direct local path with no required cloud account. Provider connections add hosted chat, image, video, embedding, and speech services without collapsing them into one opaque backend.
Working with modelsQueue follow-ups during a stream, compare several models on the same prompt, fork a complete conversation, and use the context meter and undoable compaction to keep long chats inside the model window.
Explore the pro tipsScoped API tokens expose model listing and stateless chat completions on an OpenAI-compatible /v1 surface. Stateful chats, tools, and knowledge remain on Libre’s native event API.
See the API boundaryKnowledge, tools, prompts, and skills are separate building blocks. Compose exactly what an assistant should see, then keep side effects behind a visible decision.
A model can call built-in tools, pinned OpenAPI operations, or Streamable HTTP MCP tools across multiple rounds. Read-only work can flow; side effects pause for approval.
Explore agents and toolsHybrid BM25 and vector retrieval finds exact terms and semantic matches, with citations down to a PDF page, presentation slide, spreadsheet sheet, or Markdown section.
Explore knowledgeAssistant profiles can bind a model, prompt, tools, lazy-loaded skills, knowledge collections, and a voice. Permissions are re-checked whenever the profile is used.
Assistant profilesThe solo path stays light. Team deployments add shared persistence, workers, coordination, and object storage—plus the controls operators need to understand quality, cost, and health.
Run locally with SQLite and an embedded worker, or use PostgreSQL, Redis, S3-compatible storage, and external workers for a shared multi-replica deployment.
Explore the platformVersioned model tariffs turn provider-reported usage into cost analytics. Alert-and-block budgets reject new interactive generations once tracked spend reaches the period limit.
Costs and budgetsCollect message feedback, run blind model battles with an Elo leaderboard, and execute saved evaluation sets through an admin-managed workflow.
Evaluation platform