Skip to main content

Local-first multi-agent runtime

One runtime, every agent operation in context.

Fabric connects conversations, agents, memory, automations, and runtime controls across desktop, terminal, web, messaging, and IDE workflows. Profiles keep each agent's configuration and memory isolated.

Install Fabric Explore desktop
  • macOS · Windows · Linux
  • Subscriptions · APIs · Ollama
  • Apache-2.0 licensed
PROFILE / MAIN Ready
Model route
OAuth · API · local
Memory
Profile-scoped
Skills
Loaded on demand
Control
Explicit local policy
Fabric coreshared state · shared context · shared control
  • Desktop
  • Terminal
  • Web
  • Messaging
  • IDE / API
Models
Route subscriptions, APIs, and local inference per profile.
Memory
Carry durable context across sessions without rebuilding it.
Skills
Load specialized playbooks only when the work needs them.
Work
Schedule, delegate, and continue long-running tasks.
Control
Keep risky actions behind visible, explicit approvals.

One core / every surface

Start anywhere. Keep the same agent.

Every surface can use the selected Fabric profile, so model routes, memory, skills, and session history stay aligned without turning profiles into team or tenant boundaries.

  1. 01

    Desktop

    A native workspace for chat, projects, previews, providers, memory, skills, and operations.

    Open guide
  2. 02

    Terminal

    CLI and TUI workflows with real tools, streaming progress, approvals, and scriptable output.

    Open guide
  3. 03

    Messaging

    Long-running conversations across Telegram, Discord, Slack, Signal, email, and more.

    Open guide
  4. 04

    Web

    A coordinated Workspace for agent operations and Admin console for runtime control.

    Open guide
  5. 05

    IDE + API

    ACP, MCP, and API surfaces that bring the same agent into editors and applications.

    Open guide
  6. 06

    Host it

    Run the same agent always-on: install locally, self-host on a small VPS, or follow the managed and shared-compute designs.

    Open guide

Choose your model route

Use the account you have—or keep it local.

Personal subscriptionsChatGPTConnect a personal OpenAI subscription without pasting a long-lived API key.xAI subscriptionGrokConnect SuperGrok or X Premium+ with the supported OAuth flow.API providersBring your keyChoose direct providers, routing services, or any compatible endpoint.Local inferenceOllama nativeKeep model inference on your machine and verify the local-AI boundary.

The real interfaces

One profile. Three ways to work.

These are live Fabric surfaces running against the same local profile—not concept art or staged mockups. The web Workspace/Admin foundation is available now; hosted tenancy, enforced role access, and durable enterprise ledgers remain staged backend work. See the delivery map.

Fabric TUI showing command help and session status inside the Fabric Light dashboard

Terminal UI

The full agent in your terminal.

Commands, skills, status, approvals, and live tool work without leaving the keyboard.

See how it works
Fabric web dashboard showing a connected local Ollama model

Web dashboard

Operations and control, coordinated.

Move between Workspace operations and Admin controls without leaving the local runtime.

See how it works

Compound the work

Memory that follows. Skills that deepen.

Fabric is built to improve the next session, not just finish the current prompt. Durable memory keeps useful context; skills turn repeatable methods into reusable capability.

01

Profile memory

Keep user context, project facts, and searchable session history within the selected profile.

Understand memory
02

Progressive skills

Discover and load focused instructions only when relevant, preserving a stable conversation core.

Explore skills

Install from the public repository

Your first Fabric session is one command away.

Install on macOS, Linux, or WSL2, choose a model route, and verify the environment before running real work.

Follow the quickstart View source 
curl -fsSL https://raw.githubusercontent.com/ObliviousOdin/fabric/main/scripts/install.sh | bash
fabric model
fabric status --deep