monad.ai `2.2.0`

Active execution agents inside semantic namespaces. Current status: monad.ai v2.2.0 — NRP chemistry frozen at nrp-chemistry-v0.1 (2026-05-08).

Includes full NRP mesh stack:

Phase 1–8 complete: namespace discovery, selector routing, production scoring, decision introspection, decision logs, continuous reward, low-margin exploration, patch bay
KDF domain separation: SEED env var → deterministic Ed25519 keypair via HKDF — same (who, secret) = same monad identity everywhere
monads proxy browser gateway: PAC file on port 8160, routes name.monad to running monads without DNS changes
Mesh announce: POST /.mesh/announce incoming + MONAD_SURFACE_URL outgoing heartbeat
Scope-chain routing: monad[frank] compound → rootspace → 404 fallback

Start here for implementation details:

What monad.ai does

Every HTTP request carries a Host header. The monad resolves it into a namespace and routes into the correct kernel branch:

GET /profile/name
Host: suign.cleaker.me
→ me://suign.cleaker.me:read/profile/name
→ reads profile.name from suign.cleaker.me kernel

POST /
Host: suign.cleaker.me
{ "expression": "profile.name", "value": "Sui" }
→ writes into namespace memory log (append-only, hash-chained)

The storage is the kernel. No database, no parallel ledger, no second truth.

monad[name] — scope-chain routing

me://suign.cleaker.me:read/monad[frank]/projects/x

Resolves via fallback chain:

frank @ suign.cleaker.me   (compound — exact match)
frank @ cleaker.me         (rootspace — fallback)
404

The bridge strips monad[frank] before proxying. Frank’s endpoint receives /projects/x.

Mesh — announce and discovery

Outgoing announce (MONAD_SURFACE_URL)

MONAD_SURFACE_URL=https://cleaker.me monads
# POST /.mesh/announce to cleaker.me on startup + every 30s

Incoming announce

POST /.mesh/announce
{
  "monad_id": "monad:abc123",
  "name": "frank",
  "namespace": "suign.cleaker.me",
  "endpoint": "http://raspberry.local:8161"
}

Throttled at 10s minimum between accepted registrations from the same monad_id. Entries expire after 5 minutes without heartbeat.

How namespace resolution works

The Host header determines the namespace. A monad needs no static config — every request carries its own namespace context.

Host: cleaker.me          → namespace: cleaker.me
Host: suign.cleaker.me    → namespace: suign.cleaker.me
Host: mexicoencuesta.com  → namespace: mexicoencuesta.com

One daemon can serve any number of namespaces. NetGet routes real domains to the monad port.

The kernel is the storage

Every write is a memory event in an append-only, hash-chained log:

memoryHash → prevMemoryHash → path → operator → value → timestamp → namespace

The /blocks endpoint projects that log. On restart, the daemon hydrates from the DiskStore. No migrations needed.

Identity and claims

Before writes are accepted, the namespace must be claimed. A claim anchors the cryptographic identity derived by this.me to the namespace:

POST /  { "operation": "claim", "secret": "...", "proof": { ... } }
POST /  { "operation": "open",  "secret": "...", "identityHash": "..." }

The daemon verifies the proof. It never holds the seed. The seed never leaves the client.

Environment variables

Running locally

npm install && npm run build
SEED="your-seed" node dist/src/index.js

curl http://localhost:8161/__bootstrap
curl http://localhost:8161/profile/name

The stack

this.me    → sovereign kernel. (who, secret) → seed → identity + tree.
cleaker    → resolver. projects .me into a namespace surface.
monad.ai   → daemon. runs the kernel, exposes HTTP, registers on mesh.

The namespace is not storage. The namespace is chemistry. It is the surface where identities react and compounds form.

monad.ai 2.2.0