Open Source & Self-Hosted

Libre WebUI Make whatever comes next.

Give the models you choose governed tools and cited knowledge, then work together through chat, channels, or turn-based voice. Generate and edit media in a workspace you self-host.

Try Libre Install

Apache 2.0 Self-hosted Tools off by default Private resources by default

Libre / Motion 01
The horizon stays open. What comes next is yours to make.

New in 0.26

Models can act. You stay in control.

Connect supported local or cloud models to tools, cited knowledge, and reusable skills. Libre shows every call, runs it under your identity, and pauses before anything changes state.

Explore everything in 0.26 Install

Libre WebUI 0.26 showing tool calls, cited sources, notifications, provider usage, calendar, and evaluations in one self-hosted workspace. — ToolsCitationsTeamsGovernance

Act with oversight

Native multi-round tools, OpenAPI services, and Streamable HTTP MCP servers—with explicit approval before side effects.
Read the guide
Know the source

Hybrid document retrieval finds exact terms and meaning, then cites the page, slide, sheet, or section behind the answer.
Read the guide
Reuse good work

Versioned prompts, lazy-loaded skills with companion files, and assistant profiles that bind tools, knowledge, and voice.
Read the guide
Work as a team

Channels, direct messages, shared resources, calendars, reminders, and notifications keep people and models in context.
Read the guide
Speak and refine

Hands-free turn-based voice, consent-aware saved voices, image editing, mask inpainting, compositing, audio, and video.
Read the guide
Operate with evidence

Versioned tariffs, budgets, model evaluations, structured logs, opt-in OpenTelemetry, and a release-gated team profile.
Read the guide

01 Ideas take form

A thought can become an answer, an image, or an action.

Create beyond text, work from evidence, and let supported models use tools without giving up the approval boundary.

Imagine.

Generate, edit, inpaint, and composite images beside the conversation that asked for them.

Explore

Build.

Turn a sentence into sandboxed HTML, React, SVG, Mermaid, JSON, or code you can inspect and run.

Explore

Know.

Search documents by exact term and meaning, then follow citations back to the page, slide, sheet, or section.

Explore

Act.

Let supported models call built-in, OpenAPI, or HTTP MCP tools. Anything that changes state waits for approval.

Explore

03Reusable assistants

Turn a good way of working into something reusable.

Build an assistant profile from the model, instructions, tools, knowledge, and voice the work needs. Every bound resource is re-authorized when someone uses it.

Explore agents and tools

01ModelBound

02PromptBound

03ToolsBound

04SkillsBound

05KnowledgeBound

06VoiceBound

Prompts

Slash commands with typed variables, versions, rollback, import, and sharing.

Skills

Instructions and companion files loaded only when the model needs them.

◎

Profiles

Compose model, prompt, tools, skills, knowledge collections, and voice.

04People and models

Bring the team into the same context.

Collaboration lives on the same private-by-default resource model as the rest of Libre. Membership and grants—not a global admin shortcut—decide who can read content.

Explore team collaboration

Channels with models inside

Public and private rooms, direct messages, threads, reactions, files, and @model replies that run under the asking member’s own access.

Share without flattening access

Grant chats, notes, knowledge, prompts, skills, personas, calendars, and tool servers to users or groups. Revocation applies on the next request.

Time and work stay connected

Color-coded shared calendars, recurring events, reminders, ICS import and export, plus scheduled AI runs that arrive as ordinary chats.

Events reach the right person

A durable notification inbox covers mentions, messages, shares, reminders, and failures. Signed, redacted webhooks connect external systems.

05Built to operate

Know what it costs, how it performs, and where it runs.

Libre’s operations story is part of the product: access, spend, quality, telemetry, recovery, and the path from one process to a release-tested team deployment.

Explore the platform

Cost

Price real usage.

Versioned tariffs price provider-reported tokens and media units. Budgets can observe, alert, or block new interactive generation.

Tariffs / budgets / CSV

Quality

Compare what works.

Collect structured reply feedback, run blind model arena matches, and repeat saved evaluation sets across models and providers.

Feedback / Elo / eval sets

Operations

See and recover the system.

Structured logs, correlation IDs, opt-in OTLP export, signed encrypted backups, and durable jobs make operations inspectable.

JSON logs / OTLP / recovery

Scale

Move from solo to team.

Start with SQLite on one machine. The documented team profile combines PostgreSQL, PGVector, S3-compatible blobs, Redis, and external workers.

3+ replicas / release-gated drill

OpenTelemetry export is opt-in. Hard budgets govern new interactive generation and deliberately fail open; scheduled automations remain independently controllable.

06 Your workspace

Choose the intelligence. Keep the workspace.

Use local models for private work. Reach for a cloud provider when it earns a place in the process. Every route stays visible, and every one of them stays your decision.

Explore connections

Ollama OpenAI Anthropic Google Mistral Groq OpenRouter Hugging Face ComfyUI

01 / Local

Private when it matters.

Run Ollama on hardware you own, and a local conversation never leaves it.

02 / Optional

Connected when you choose.

Add a remote provider for the work that needs one — never as a silent default.

03 / Extensible

Open by design.

Describe any OpenAI-compatible service in JSON, or talk to a coding agent already installed on the machine.

07 Begin

Start with one line.

No vendor account, hosted seat, or trial. Run one command, create your local administrator, and the workspace is yours.

npx libre-webui@latest

The shortest path from a terminal to a working workspace. NODE 22.22+

Opens at localhost:8080 No vendor account required. Add Ollama for local models or connect a provider you choose.

All downloads↗︎ Read Documentation↗︎ GitHub↗︎

08 Extend

Bring your models, tools, and clients.

Provider plugins connect inference and media. OpenAPI and Streamable HTTP MCP servers add tools. Existing OpenAI clients can use Libre’s stateless chat API.

Build a connection Browse plugins ↗︎

my-studio.json Plugin

{
  "id": "my-studio",
  "type": "multi",
  "capabilities": [
    "chat", "image", "voice"
  ],
  "endpoint":
    "http://localhost:8000/v1"
}

Connected localhost:8000

Available capabilities 05

01 Conversation Available

02 Image Available

03 Voice Available

04 Documents Available

05 Agents Available

Libre WebUI

Start in one line.Leave whenever you want.

No vendor account, hosted seat, or trial. Run it on hardware you control — and if this project ever stops serving you, your installation and your fork do not need our permission.

Try Libre Install

Apache 2.0 No Vendor Telemetry Self-Hosted

Libre WebUI Make whatever comes next.

Models can act. You stay in control.

Act with oversight

Know the source

Reuse good work

Work as a team

Speak and refine

Operate with evidence

A thought can become an answer, an image, or an action.

Imagine.

Build.

Know.

Act.

Give an agent a place to work.

Turn a good way of working into something reusable.

Bring the team into the same context.

Channels with models inside

Share without flattening access

Time and work stay connected

Events reach the right person

Know what it costs, how it performs, and where it runs.

Price real usage.

Compare what works.

See and recover the system.

Move from solo to team.

Choose the intelligence. Keep the workspace.

Private when it matters.

Connected when you choose.

Open by design.

Start with one line.

Bring your models, tools, and clients.

Model and media providers

OpenAPI and HTTP MCP tools

OpenAI-compatible clients

Start in one line.Leave whenever you want.