AI agents for the semiconductor industry. Barbot is a multi-product platform that turns the technical grind of chip work — cross-referencing, selection support, knowledge lookup — into AI agents. One platform, one credit balance, every product.
Built for chip vendors and distributors: the buyers are FAE / sales / customer engineers whose core job is selection support.
| Product | What it does |
|---|---|
| Chip P2P | AI reads datasheets and finds pin-to-pin alternatives in minutes — parameter-level diffs with datasheet page-anchored provenance, a substitution verdict tag (pin2pin / functional / not-substitutable), editable reports, one-click blog publish. |
| AI FAE | A 24/7 AI Field Application Engineer for your website — answers technical questions and grounds substitution answers in the live chip catalog, captures leads, hands off to human support. |
| EE Block Diagram | Describe an electronic system, get a functional block diagram — exact-label structured SVG (downloadable vector) or a datasheet-style AI rendering. |
| Harvey Agents | The heavy-duty agent workbench — connect data sources and MCP tools, run long batch tasks (datasheet crawling, BOM matching), keep every session organized. |
All products share one account, one credit pool, and one billing model: pay per use, with cache hits free.
Barbot is built on a headless SaaS engine (payments, credits, subscriptions, auth, RBAC, i18n) with the chip-vendor product surface built on top. The platform (control plane) runs the web app + billing; the Harvey agent workbench (execution plane) runs the heavy agent workloads and plugs into the platform through a thin token / usage / balance contract.
src/
├── core/ # Infrastructure (db, auth, payment, email, storage, ai, i18n)
├── modules/
│ ├── chips/ # chip catalog + shared agent tools
│ ├── chip-compare/ # datasheet compare pipeline, EE diagram engines
│ ├── ai-support/ # AI FAE chatbot + knowledge base
│ ├── agent-gateway/ # platform contract: tokens, usage, device login, membership
│ └── credits · rbac · payment · subscriptions · apikeys · ai-tasks
├── routes/ # File-based routes (pages + /api server routes)
│ ├── compare/ diagram/ chips/ # Chip P2P + EE diagram tools
│ └── api/agent/ api/auth/desktop/ api/internal/ # agent + membership contract
├── config/ # Environment, DB schema templates (sqlite/postgres/mysql)
└── components/ blocks/ lib/
sdk/barbot-agent-sdk.mjs # zero-dependency client for external agent executors
deploy/ # production docker-compose stack + runbook
pnpm install
cp .env.example .env.development # set VITE_APP_URL, DATABASE_*, AUTH_SECRET
pnpm db:push
pnpm rbac:init --admin-email=admin@example.com --admin-password=your-password
pnpm dev # http://localhost:3000Then at /admin/settings → AI: enter the DeepSeek API key and model; import a
chip catalog CSV at /admin/chips. AGENTS.md has the full developer guide.
Production deploy (VPS · PostgreSQL · Caddy auto-HTTPS · platform + Harvey) is
scripted in deploy/ — see deploy/README.md
for the full runbook.
- TanStack Start (Vite 8 + nitro, React 19, TypeScript strict), file-based routing
- TanStack Query / Form / Table; shadcn/ui v4 (Tailwind CSS 4)
- better-auth + Drizzle ORM (SQLite / PostgreSQL / MySQL)
- Paraglide JS for i18n (English + Chinese)
- LLM: DeepSeek (Anthropic- and OpenAI-compatible endpoints); gpt-image via relay
- All business logic self-contained — no external SaaS for payments/credits/auth
| Command | Description |
|---|---|
pnpm dev |
Vite dev server (port 3000) |
pnpm build |
Production build |
pnpm start |
Run the production server |
pnpm agent:worker |
Agent-task worker (add WORKER_LOOP=1 to run resident) |
pnpm db:setup / db:push / db:generate / db:migrate |
Schema tooling |
pnpm rbac:init / rbac:assign |
RBAC bootstrap + role assignment |
Proprietary. See LICENSE.