Day to day

The install is done and an agent answers. This is what the rest of the week looks like.

Two interfaces, and they are for different things

The dashboard is for anything an agent owns — its personality, its model, its skills, who may talk to it, what it has scheduled. It is a browser tab at http://127.0.0.1:18254/dashboard/.

The CLI is for the daemon — is it up, what is queued, what is in the audit log, restart this gateway. It works over SSH and against a remote daemon.

Neither is a lesser copy of the other. The setup wizard and nouride setup drive the same code; the skills page and nouride skills write the same file.

Neither of them is how an agent works. An agent acts through tools — each one typed, individually gated, and audited by name. It does not shell out to nouride, and the design makes that unattractive on purpose: the control API needs a token, and handing an agent that token would collapse every per-tool approval into a single approved exec.

The things you will do

	Where	Page
Talk to an agent	any chat app, the dashboard, or nouride chat	Talking to an agent
Answer “may I run this?”	the chat it asked in, or the dashboard	Approvals
Change how an agent behaves	dashboard → Agents → Files	Persona packs
Switch models	dashboard → Agents → Model	Provider profiles
Decide what it keeps learning	dashboard → Skills	Skills & learning
Let someone in, or keep them out	dashboard → Agents → Access	Who may talk to it
Make something happen at 9am	dashboard → Schedule, or nouride cron	Scheduled work
Check on it	nouride status, nouride doctor	doctor
Back it up, upgrade it	tar over one directory	Maintenance

The one directory

One install is one directory. This is the single most useful thing to know about operating it.

/srv/nouride/           (Linux)   ~/nouride/   (macOS)   ./nouride/   (Docker volume)
├── nouride                the binary
├── dashboard/             the web UI it serves
├── skills/                skills that ship with the daemon
├── wa-bridge/             the WhatsApp bridge, its own executable
├── .env                   credentials and per-host values (0600)
└── .nouride/
    ├── config.toml        everything structural
    ├── data/              SQLite, secrets, attachments, backups
    ├── agents/            personas and agent packs
    │   └── skills/        the skill pool — every agent draws from it
    └── workspace/         what the agents produce

Back up .nouride/. Everything else is a download.

Five commands worth memorising

nouride status          # agents, gateways, queue — one screen
nouride doctor          # diagnose a setup that is not working
nouride logs -f         # live log; works with no journal, and against a remote daemon
nouride chat            # talk to an agent from the terminal, no bot token
nouride approvals       # what an agent is waiting on you for

They find their own install, so they work from any directory on the install’s machine — see the note about working directories in troubleshooting if one of them reports no daemon while one is running.

Driving a remote daemon

Every command except start talks to a running daemon over the control API. Point it somewhere else with two environment variables and the whole CLI works against that box:

export NOURIDE_URL=https://nouride.example.com
export NOURIDE_TOKEN=<the contents of its data/.control-token>
nouride status

What runs without you

Some things happen whether or not anyone is watching, and it is worth knowing which:

• Scheduled jobs. A cron turn that needs an approval is auto-approved by default and the decision is recorded in the audit log against the job — because a 3am job that stops at a prompt nobody will see has not run at all. The hardline classes still ask, and still fail closed.
• Reconnects. A gateway that drops backs off and retries. One that gives up entirely raises an alert, if you configured somewhere for alerts to go.
• Retention sweeps. Old audit rows, old job runs and old attachments age out on the schedule in [retention].
• Persona backups. Dated copies of the agent packs, under data/backups/agents/. Kilobytes of Markdown, and the only thing in the install that cannot be rebuilt.
• Nothing else. An agent does not act unless a message, a job or a webhook wakes it.