Stop building platform plumbing

Without Fabric

• Build auth middleware
• Build multi-tenant org model
• Build RBAC from scratch
• Build job queue + retries
• Build provider abstraction
• Build SSE event system
• Build cost tracking
• Build webhook delivery
• ~3 months before you ship a feature

With Fabric

just run

What’s inside

Identity

Orgs, teams, memberships, invitations, service accounts, API keys. Your app reads from Fabric — never owns user state.

Auth

Four roles (Owner/Admin/Member/Viewer), effective permission computation, batch checks. Route-level enforcement built in.

Workflows

Python DAGs with .then(), .fork(), retries, timeouts, shared context. Write tasks as async functions, run them on N workers.

AI Routing

OpenAI, Anthropic, Gemini, Ollama, Whisper, fal.ai, ComfyUI. Tier-based selection, cost-aware fallback, health checks, streaming.

Events

SSE, WebSocket, and HMAC-signed webhooks. Every state transition. Per-run replay. No polling.

Runtimes

Native (Rust), Tool (ffmpeg, yt-dlp), Provider (AI), Wasm (sandboxed plugins). Same semantic operation, four execution strategies.

How it’s built

Two planes. The control plane owns all state — identity, tenancy, RBAC, workflows, providers, cost, audit. HTTP on :3001, gRPC on :3002, SSE and WebSocket for real-time.

The execution plane is stateless. Workers claim steps with SELECT … FOR UPDATE SKIP LOCKED, execute them, report results. Scale by adding more workers.

Everything goes through Postgres. No Redis, no message broker, no distributed consensus. The database is the queue.

Control Plane:3001 HTTP · :3002 gRPC · SSE/WS

Postgres

Workers (N)Native · Tool · Provider · Wasm

Try it

brew install just protobuf
just infra-up && cp .env.example .env
just run          # control plane
just executor     # workers

Python

from fabric_platform import FabricClient
fabric = FabricClient()

# topic in, production video out
run_id = fabric.run_workflow("global/ai-shorts", context={
    "topic": "Why sleep is a superpower",
})

result = fabric.wait_for_run(run_id)
print(result["output"]["final_video_path"])

TypeScript

import { FabricClient } from "@fabric-platform/sdk";
const fabric = new FabricClient();

const run = await fabric.workflows.runs.submitRun({
  workflowSlug: "global/ai-shorts",
  input: { topic: "Why sleep is a superpower" },
});

for await (const event of fabric.workflows.runs.streamRunEvents(run.id)) {
  console.log(event.kind);
}

curl

curl http://localhost:3001/healthz

curl -X POST http://localhost:3001/v1/workflows/global%2Fai-shorts/runs \
  -H 'content-type: application/json' \
  -d '{"context":{"topic":"Why sleep is a superpower"}}'

Built-in workflows

Fabric ships workflows you can run immediately — or use as starting points for your own.

AI ShortsTopic in, 45-second video out. Script, actor, voiceover, b-roll, music, subtitles — 6-way parallel.Problem IntelligenceScrape Reddit/HN/YouTube/Twitter. Extract problems, cluster, score, generate startup ideas and specs.Clip GeneratorDrop a YouTube URL. Get vertical clips with face tracking, subtitles, hook overlays, and effects.Research → VideoDeep research a topic, generate hooks, fan out N parallel video renders. One command.

InstallationWorkflow authoringPython SDKWorkflow SDKAPI referenceModel config