Serve mode starts a background daemon that keeps one model loaded, so repeated runs skip the weight loading.
neurobrix serve --model <name>
Expected result: the daemon starts and loads the model. If one is already running: [Serve] Daemon already running (PID n) and exit code 1.
neurobrix run --prompt "A sunset over mountains"
Expected result: output from the loaded model. Without --model, run takes the model from the running daemon; with no daemon and no --model it stops with ERROR: --model is required when no daemon is running.
neurobrix stop
Expected result: the daemon shuts down (socket shutdown, then SIGTERM, then SIGKILL) and stale files are cleaned up.
The daemon handles one task at a time. A cold run while it is up is refused.
For language models:
neurobrix chat
chat connects to a running daemon. Options: --max-tokens, --temperature, --repetition-penalty, --agent. Inside the session: /new, /context, /status, /quit.
| Flag | Meaning |
|---|---|
--model |
Required |
--hardware |
Hardware profile id, see Hardware Profiles |
--timeout |
Idle seconds before the daemon shuts down itself; default 1800 (30 minutes) |
--foreground |
Do not detach |
--compiled, --sequential, --triton, --triton-sequential |
Execution mode, at most one; see Inference Guide |
In ~/.neurobrix: daemon.sock plus a .pid and a .log file (on Windows a TCP port file replaces the socket). When something goes wrong, the .log file is the first place to look; see Troubleshooting.
Which families benefit from the warm flow, and the load time saved per model: not documented yet.
Next: CLI Reference
© 2025–2026 Hocine Benkelaya. NeuroBrix engine released under the Apache License 2.0.
Developed by Hocine Benkelaya · Supported and funded by Neural Network Holding Ltd · Commercial distribution by WizWorks OÜ
Documentation source: GitLab