Skip to content

Shared Daemon

Nexus = filesystem/context plane.

Choose this path when Nexus needs to run as a service instead of as an in-process library.

What Changes

  • You start nexusd as the server process.
  • Remote SDK clients use the remote profile.
  • The remote SDK path depends on gRPC as well as the HTTP server URL.

Server

Run the daemon directly (supported copy-paste workflow):

export NEXUS_GRPC_PORT=2126
nexusd --profile full --host 127.0.0.1 --port 2026 \
  --data-dir ./nexus-data --auth-type static --api-key dev-key

The managed stack (FULL profile: PostgreSQL + Dragonfly + the Nexus server; Zoekt is optional and separately run) is nexus init --preset shared then nexus up, then eval $(nexus env).

Known issue (Bug B, tracked): do not chain these with && — nexus up --preset shared currently exits non-zero because its health gate waits on a zoekt service the preset does not start, so a &&-chained eval $(nexus env) would never run. The hub itself boots and serves correctly; only the nexus up wrapper's exit status is wrong. Prefer the direct nexusd command above until the out-of-scope nexus up fix lands. See FULL deployment profile ("Known issue").

minimal is not a deployment profile. Valid profiles: embedded, lite, sandbox, full, cloud, cluster, remote (and remote cannot run as a daemon). See FULL deployment profile.

Client

from nexus.sdk import connect

nx = connect(
    config={
        "profile": "remote",
        "url": "http://127.0.0.1:2026",
        "api_key": "dev-key",
    }
)

Set NEXUS_GRPC_PORT in the client environment if the server is not using the default gRPC port expected by the SDK.

Trust Notes

  • The HTTP port alone is not enough for the current remote SDK path.
  • Documented remote access should always mention the gRPC requirement explicitly.
  • If you only need local development, start with the Local SDK path.