Skip to main content

Cluster

--name and --url are only for clusters. The coding-agent flags are only for claude, codex, opencode, and cursor targets. Mixing them is an error. Cursor supports session reporting only. connect cursor and disconnect cursor manage its reporting. Routing flags such as --agent and --config are rejected for Cursor.

Register a cluster

connect checks the cluster’s /health first and stops if it doesn’t answer. It then saves the cluster in ~/.nasiko/config.json, makes it active, and signs you in. Sign-in is skipped when you already have a live session for that cluster, or when there is no terminal to prompt on. A failed sign-in doesn’t undo the registration: run nasiko auth login later. Once you’re signed in, connect asks which coding agents on this machine should report sessions to the cluster. It lists each detected agent that has no reporting yet, and each agent that reports to a different cluster or account, labelled reports sessions to <cluster>; select to move it here. Nothing is selected to start with. Select none to change nothing. An agent you leave unselected isn’t offered again for that cluster and account. It is still offered when you connect to a different cluster. Connecting the agent later, with nasiko connect <agent>, clears the decline everywhere. Removing an agent’s reporting with nasiko disconnect counts as a decline for the cluster it reported to. When at least one agent reports here, connect asks whether to also send the sessions you ran before connecting, newest first. Sessions already sent are skipped.

Connect a coding agent

A coding agent can get three things from Nasiko: Omit the target to pick from the coding agents installed on this machine. On a terminal, connect asks about each part, and flags answer the same questions in scripts. Without a terminal, a new routing connection gets MCP connectors, and session reporting stays off unless you pass --reporting, --content, or --no-content. Reporting always syncs usage: tokens, cost, latency, models, tools, and subagents. Chat history (prompts, responses, tool output) is synced only to a cluster that keeps it, and only after you agree or pass --content. A cluster that doesn’t keep chat history gets usage only. Some coding agents need one more step after reporting is turned on. For Codex, run /hooks in Codex and trust the new Nasiko hooks. For OpenCode, restart it so it loads the plugin.

Disconnect a coding agent

On a terminal, omit the target to pick from the coding agents that are connected. Without a terminal, name the coding agent. Without --routing or --reporting and without a terminal, disconnect removes LLM routing only. For Cursor, which has no routing, it removes session reporting. Past sessions stay on the server.

Status

nasiko status prints the active cluster’s health (Postgres, Redis, orchestrator) and counts of running agents, containers, and pending builds. It then prints one row per coding agent on this machine:
INSTALLED is yes when the coding agent’s binary is on your PATH or its config directory exists. REPORTING says whether sessions sync usage only, or usage and chat history. The command exits with code 1 when the cluster is unhealthy, so nasiko status && nasiko deploy . stops on an unhealthy cluster. The coding-agent overview prints even when the cluster can’t be reached. --json prints only the cluster status as one JSON object: cluster, url, registry_url, healthy, health (postgres, redis, orchestrator, each ok or fail), and metrics (agents_running, agents_total, containers_total, builds_pending). It can’t be combined with a coding-agent target. nasiko status <agent> shows that coding agent’s LLM routing and session reporting state, and where its reporting hook lives.

Auth

With no flag, auth login signs in through the cluster’s SSO in your browser when the cluster has SSO configured. Otherwise it prompts for a username and password. --password always prompts for a username and password. --sso and --password can’t be combined.
Plans: Free, Individual, Pro, and Enterprise. Self-hosted: Enterprise license.
Browser sign-in through SSO (nasiko auth login --sso, and the automatic SSO choice) needs a server with an identity provider configured: a Nasiko Cloud workspace, or a self-hosted Enterprise server. Against a self-hosted open-source server, auth login uses a username and password, and --sso fails with a message to use --password. See SSO and SCIM.
With --sso, the CLI shows a code and opens a browser page. Approve the code there while signed in through your identity provider. The CLI waits until you approve, then saves the session. If the cluster has no SSO configured, --sso fails and tells you to use --password. After auth login, nasiko connect <url>, or nasiko use, the CLI offers to turn on session reporting for the coding agents it finds on this machine, as described in Register a cluster. Without a terminal, it prints the command to run instead: nasiko connect <agent> --no-routing --reporting. See Discover and connect.

Context budget

See Reduce cost.

Diagnostics and shell setup

doctor

Run doctor when a command fails partway through, or before a demo. It checks, in order:
  1. Docker (or Podman) answers.
  2. The active cluster is reachable.
  3. You are signed in and the saved token hasn’t expired.
  4. An artifact registry URL is set. This is advisory: most commands don’t need it.
Each check waits at most 5 seconds. doctor only reads state. It never changes config, starts anything, or signs you in.
The fix for each problem is printed under it. doctor exits with code 1 when it finds a problem. The registry warning doesn’t count as one.

completions and man

Completion scripts and the man page leave out hidden commands.