@arcforge/axond

The machine-wide daemon for the Axon agent runtime.

One process per user, per machine, owning everything no single agent can own. Not an inference server — inference is its first tenant, not its purpose. Four domains:

• machine — the GPU. Two agents deciding independently that 6GB is free is how both take it, so one process arbitrates.
• agents — every agent running here, in a registry that outlives the terminal that started one.
• models — one resident copy of a weight, however many agents hold it.
• schedule — boot-time agents and cron-style wakeups.

Axond() is the server; AxonDaemon() is the client handle, and both expose the same four domains. The client is the SDK: daemon.agents.at(id) returns an instance handle you can talk to, not a record you look things up in.

npm install @arcforge/axond

Most people never install this directly. It arrives with the Axon CLI, which keeps the daemon running for you — axon daemon status to see it.

import { AxonDaemon } from "@arcforge/axond"

const daemon = AxonDaemon()

await daemon.machine.state()   // hardware, budget, what is held
await daemon.agents.list()     // every agent on this machine
await daemon.models.resident() // what is loaded right now

Documentation

Full documentation is at axon.arclabs.it/docs/v2.

• What is Axon?
• The CLI

License

Proprietary. © ArcLabs