Skip to content

Phase 1: FAGS MVP

Goal

Ship the minimum viable FreeAgentStore: a working store with 3-5 browser-based AI agents, a publish flow, and the foundation for growth.

GitHub org

  • Org: FreeAgentStore (https://github.com/FreeAgentStore)
  • Primary repo: FreeAgentStore/platform (monorepo)

Directory structure

~/dev/stores/fags/
├── platform/                    # Core monorepo (vendored from FAS)
│   ├── packages/
│   │   ├── sdk/                # @freeagentstore/sdk
│   │   ├── cli/                # @freeagentstore/cli (fags binary)
│   │   ├── compliance/         # Agent-specific compliance checks
│   │   ├── backend/            # CF Worker (api.freeagentstore.online)
│   │   └── quality/            # Agent quality reporter
│   ├── package.json
│   └── pnpm-workspace.yaml
│
├── freeagentstore/             # Store site (static HTML)
│   ├── index.html
│   ├── registry.json
│   └── build.js
│
├── host/                        # Host Worker (*.freeagentstore.online → R2)
├── admin/                       # Admin Worker (provisioning)
├── publisher/                   # Publisher Worker (self-service)
├── agent/                       # VibeCode agent builder
├── create/                      # VibeCode UI (React)
├── console/                     # Creator console
│
├── templates/
│   ├── template-agent-tts/      # Kokoro TTS template
│   ├── template-agent-whisper/  # Whisper STT template
│   ├── template-agent-vision/   # Image processing template
│   ├── template-agent-llm/      # Text generation template
│   └── template-agent-tools/    # Utility tools template
│
├── agents/                      # Published agents (cloned repos)
│   ├── audiobook/              # First agent (port from bepub)
│   ├── transcriber/            # Whisper-based transcription
│   └── bg-remover/             # Background removal
│
└── CLAUDE.md

Step-by-step build order

Step 1: Platform monorepo scaffold

  1. Create FreeAgentStore/platform repo on GitHub
  2. Vendor FAS platform structure (copy, don't depend)
  3. Rename all freeappstore → freeagentstore, fas → fags
  4. Set up pnpm workspace

Step 2: SDK (@freeagentstore/sdk)

Start from FAS SDK, add agent-specific primitives:

// New exports (agent-specific)
export { useModel } from './model';          // Model download + cache + inference
export { useWorkerInference } from './worker'; // Web Worker orchestration
export { useOllama } from './ollama';          // Local LLM detection + API
export { useModelCache } from './cache';       // Shared model caching
export { useResultStore } from './results';    // IndexedDB result persistence

// Inherited from FAS SDK
export { Auth, Kv, Rooms, ApiProxy, Roles } from './core';

Step 3: Templates

Each template is a complete, buildable agent app:

template-agent-tts (port from bepub): - Kokoro TTS pre-wired - Web Worker with WebGPU/WASM fallback - Model download progress UI - Audio player component - Text input with chapter splitting

template-agent-whisper: - Whisper small pre-wired - Audio file drop zone - Real-time transcription progress - Timestamp display - Copy/download transcript

template-agent-vision: - RMBG/SAM pre-wired - Image drop zone - Canvas-based result display - Download processed image

Step 4: Infrastructure

  1. Doppler project: fags (auto-sync to GitHub org)
  2. R2 bucket: fags-agents
  3. D1 database: fags
  4. Host Worker: deploy to *.freeagentstore.online
  5. API Worker: deploy to api.freeagentstore.online
  6. DNS: Cloudflare zone for freeagentstore.online

Step 5: Compliance checks

Agent-specific checks (replace/extend FAS checks):

Check Rule
no-cloud-inference Free agents must not call cloud AI APIs (OpenAI, Anthropic, etc.)
web-worker-inference AI inference must run in Web Worker (not main thread)
model-size-warning Warn if total model size > 2GB
privacy-audit No data sent to external services (except model CDN download)
webgpu-fallback Must have WASM fallback if using WebGPU
model-cache Must use Cache Storage for downloaded models
manifest-valid Agent manifest (name, model, task type) must be valid
offline-capable Should work offline after first model download
brand-fonts Same Fraunces + Manrope requirement
brand-tokens Same color token system
bundle-size App bundle < 1MB (excluding models)
license-mit MIT required on free tier

Step 6: First agents

  1. Audiobook Maker — port bepub's Kokoro TTS into template
  2. Transcriber — Whisper small, drop audio get text
  3. Background Remover — RMBG, drop image get cutout

Step 7: Store site

Static HTML, same pattern as freeappstore.online: - Agent cards with: name, description, task type, model used, download size - Category filters - Search - "Open" button (not "Install" — it's a web app)

Step 8: Publisher + Admin

Vendor from FAS, swap domains. Same provision flow: 1. Create GitHub repo in FreeAgentStore org 2. Insert R2 hosting route 3. Create custom domain + DNS 4. Add to registry.json

Definition of done (Phase 1)

  • [ ] Platform monorepo builds and tests pass
  • [ ] SDK published to npm (@freeagentstore/sdk)
  • [ ] CLI published to npm (@freeagentstore/cli)
  • [ ] 3 templates available (TTS, Whisper, Vision)
  • [ ] 3 agents live on store
  • [ ] Store site at freeagentstore.online
  • [ ] Publisher portal working
  • [ ] Admin worker provisioning
  • [ ] Host worker serving agents from R2
  • [ ] Compliance checks passing for all agents
  • [ ] Doppler + GitHub org secrets configured