Configuration
orbit.Config is bound from the modules.orbit.* subtree of your nucleus.yml
(or set directly in Go). Every field is optional — the zero value mounts a
working panel under /admin.
The tables below group the keys by what they affect. For a plain in-process
panel, only the first four groups can ever matter, and most apps set just
Mounting and the bootstrap user.
The cluster_* keys belong to the opt-in live-feed relay: they are
inert until cluster_enabled is true, and in particular no Redis is
required to run the panel — cluster_redis_url is read only by the
relay.
Mounting
Key (modules.orbit.*) | Type | Default | Description |
|---|---|---|---|
prefix | string | /admin | URL path Orbit mounts under. |
title | string | Orbit | Heading shown in the UI: the login page, the sidebar, and the browser tab. |
environment | string | — | Label shown in the UI (e.g. production). |
The bootstrap admin user
Key (modules.orbit.*) | Type | Default | Description |
|---|---|---|---|
bootstrap_username | string | — | Admin user created on first boot. |
bootstrap_email | string | — | Email for the bootstrap user. |
bootstrap_password | string | — | Password for the bootstrap user. Leave it empty to skip creating the user and provision the admin account another way, e.g. nucleus createuser. The nucleus_admin_users schema is created at mount either way, so createuser works without ever setting a bootstrap password. |
auth_database | string | app default | Database alias whose handle backs admin login and the bootstrap user — point it at a dedicated database to keep the admin user store away from application data. Only login and bootstrapping are redirected; the panel itself always reads through the application's default handle. |
Data and views
Key (modules.orbit.*) | Type | Default | Description |
|---|---|---|---|
migrations_path | string | migrations | Directory the migrations view reads. |
audit_store | string | database | Where the audit trail is kept: database writes a table the panel creates and owns, so the trail survives restarts and is shared by every replica; memory keeps the process-lifetime ring instead. An application with no database handle gets the ring either way (see Audit log). |
audit_retention_days | int | 0 | Drop trail entries older than this many days — a period, which is what a compliance window is. Applied when the panel comes up and at most hourly afterwards; zero keeps entries until the log is cleared. An operator can change the window in effect from the panel, and a restart comes back to this value. |
audit_max_size | int | 10000 | In-memory audit-log ring size. It bounds the memory store only — a trail in the database is bounded by audit_retention_days, not by a count. |
multitenant_enabled | bool | false | Confine Data Studio to the tenant the host application resolves for the request — every operation, exports and their jobs included, not only the list (see Features); a request with no resolved tenant and no default is a 403. ?tenant=<id> / ?tenant=all are accepted only from a superuser or a subject granted the tenant_switch RBAC action, and audited as tenant.override. The host's resolution must not be client-controlled (a header the client sets) for the confinement to hold. |
multitenant_default | string | — | Tenant applied when the host resolves none; without it such a request is refused unless the operator may switch tenants. |
multitenant_ids | []string | — | Known tenant IDs for the selector UI. |
row_owner_fields | map[string]string | — | Which column of each model says WHICH OPERATOR a row belongs to, keyed by model name, with "*" as the default for every model carrying the same column. It is what makes an admin:<Model>#own policy enforceable (see Access control); a #own grant on a model with no entry here is refused with a 403, never widened to every row. |
row_owner_subject | string | username | Which name of the operator the owner column holds: username or id. |
field_widgets | map[string]string | — | How a field is edited when its type cannot say: Model.Field (or Model.column) to one of json, richtext, file, image. A JSON document is inferred from the type and needs no entry. A file/image field gets an upload route that stores the bytes in the application's storage and answers with the key the form writes (see Features). |
The live feed
Key (modules.orbit.*) | Type | Default | Description |
|---|---|---|---|
live_exclude_patterns | []string | — | Path patterns excluded from the live HTTP feed — use it to keep health checks and static assets out. |
trace_url_template | string | — | External trace-explorer URL template, to deep-link each entry (supports {trace_id}). |
The live-feed relay (the cluster_* keys)
By default the live feed shows this node's traffic. Turn these keys on and the nodes of one application relay their live events to each other over Redis, so the panel on any node shows the whole set.
Key (modules.orbit.*) | Type | Default | Description |
|---|---|---|---|
cluster_enabled | bool | false | Aggregate the live feed across nodes via a Redis relay. |
cluster_redis_url | string | — | Redis URL for the relay. |
cluster_channel | string | nucleus:admin:live:v1 | Pub/sub channel for the relay. |
cluster_node_id | string | runtime id | Explicit node identifier in the relay. |
cluster_token | string | — | Shared secret to reject untrusted relay messages. |
The cluster_* keys stay inside your application process: same panel, same
binary, one shared feed. The standalone agent-and-server
fleet plane is a different, heavier option — a
dedicated observability server that application nodes stream to, with no
Redis involved. Most applications need neither.
Example
# nucleus.yml
modules:
orbit:
prefix: /admin
title: Acme Admin
environment: production
bootstrap_username: admin
bootstrap_email: admin@acme.test