Skip to content

FAGS — FreeAgentStore

URL: https://freeagentstore.online Org: https://github.com/FreeAgentStore (~52 repos) Stage: Early beta Doppler: fags

Vision

A curated store of AI-powered tools that run free in the browser. Not a directory of configs. Real apps that do real work — transcription, TTS, background removal, code review — powered by client-side AI on the user's hardware.

Core thesis: Every useful AI tool people pay $20/mo for can be rebuilt to run entirely in the browser at zero marginal cost.

Architecture

fags/
├── platform/          Monorepo (pnpm)
│   ├── packages/sdk       @freeagentstore/sdk
│   ├── packages/cli       @freeagentstore/cli (fags init/check/publish)
│   ├── packages/compliance 9 agent-specific checks
│   ├── packages/agents    48 agents (canonical source)
│   ├── workers/host       Host worker (store site, proxy, key vault)
│   └── workers/mcp        MCP server (11 tools)
├── freeagentstore/    Store HTML (static, auto-generated)
├── agents/            Legacy (1 agent, tts)
└── templates/         template-agent-tts

Three types of items

Type Count AI source Cost to user
Library ~29 Pure TS heuristics (LLM-evolved code) Free forever
Model ~14 ONNX via Transformers.js (WebGPU/WASM) Free forever
Agent ~5+ User's own API key via platform proxy User pays provider

Key systems

  • Key vault: AES-256-GCM encrypted user API keys for 6 providers (OpenAI, Anthropic, Google, Groq, OpenRouter, Together)
  • Proxy: /v1/proxy/{host}/{path} injects keys server-side
  • Store generation: registry.json -> build-index.js (cards) + build-details.js (detail pages)
  • OIDC publishing: npm packages published via GitHub Actions trusted publisher

Gaps

  • Worst test ratio (0.17) — 354 source files, 60 tests
  • SDK/CLI at 0.1.0 (never bumped)
  • Many agents are thin wrappers with zero testing
  • 9 compliance checks only (vs 20-42 in FAS/FGS/PAS)

Future direction

  • Per-agent smoke tests (import + core function check)
  • Grow model catalog (target: 20+ models)
  • Chrome Built-in AI integration (Gemini Nano, Prompt API)
  • Agent composition (chain Library + Model + Agent)
  • Evolved heuristic pipeline (user feedback -> re-evolution cycles)