ChipLab docs
The same WASM core runs in the browser, Node, MCP, npm, PyPI and the single EXE. No cloud required.
Run it
# web + API (Node 22)
node backend/server.mjs
# open http://127.0.0.1:4173
# CLI
node backend/run.mjs --compare --count 1000000
node backend/run.mjs --chip 6 --mode 1 --count 250000
# desktop app (single file)
dist\chiplab.exe --port 4173
# docker
docker compose up --build
The web app also works as a static PWA. WebGPU is optional; Canvas2D is the fallback.
Local API
GET /api/health
GET /api/chips
POST /api/run { "chip": 1, "count": 100000, "seed": 1337, "mode": 0 }
POST /api/compare { "count": 100000, "seed": 1337 }
POST /api/book { "count": 100000, "seed": 1337 }
All numbers are model numbers at 322.26 MHz. The feed is synthetic and reproducible from the seed.
MCP
node mcp/server.mjs
# tools: chiplab.chips, chiplab.run, chiplab.compare,
# chiplab.book, chiplab.formal_status
The MCP server is a thin stdio JSON-RPC wrapper over the same WASM core.
Honest limits
- Pre-silicon. Not a tapeout, not silicon-proven.
- The RTL async bypass is one clock after the ingress register, not zero clocks from the wire.
- WebGPU visualises a sample of the run; totals come from the exact parser.
- Multi-frame MoldUDP64 back-to-back flow control is still experimental (single-frame proof passes).
- No investment advice, no order routing, no live exchange connection.