version: "1.0.1" name: ella-core description: > Manage and inspect a running Ella Core private mobile network via its REST API. Use when the user asks to provision, update, list, inspect, or delete subscribers, profiles, slices, policies, data networks, radios, routes, BGP peers, NAT, operators, or users; or asks about live status, data usage, flow reports, audit logs, AMF/SMF/UPF state, PDU sessions, gNB connections, IMSIs, S-NSSAIs, DNNs, 5QI, AMBR, ARP, or any runtime telemetry of an Ella Core instance. metadata: version: 0.2.0 author: ellanetworks

Ella Core Skill

Ella Core is a 4G and 5G private mobile network packaged as a single Go binary (AMF + SMF + UPF + AUSF). This skill operates a running instance via its REST API.

Connection

Two values are required:

• Base URL — e.g. http://192.168.1.10:5000. Read from ELLA_CORE_URL if set.
• API token — string prefixed with ellacore_. Read from ELLA_CORE_TOKEN if set.

Ask the user for any value that is not in the environment or earlier in the conversation. Token minting and roles: references/auth.md.

Authenticate every request with Authorization: Bearer $ELLA_CORE_TOKEN. Exceptions (no auth required): GET /api/v1/status, GET /api/v1/metrics, GET /api/v1/openapi.yaml.

Ella Core typically serves HTTPS with a self-signed certificate. Use curl -k (or --cacert <file> if the user provides one) for https:// URLs. Plain HTTP works against h2c instances.

Calling the API

Use curl -sk (silent + skip cert verification) and pipe JSON through jq (or python3 -m json.tool):

curl -sk -H "Authorization: Bearer $ELLA_CORE_TOKEN" \
  "$ELLA_CORE_URL/api/v1/subscribers?per_page=100" | jq

Retrieval-first. Before calling any endpoint not already used in this session, check references/examples.md for a recipe. If the operation isn't covered, fetch the OpenAPI spec — the source of truth, never guess paths or shapes:

curl -s "$ELLA_CORE_URL/api/v1/openapi.yaml"

Fetch once per session and reuse. Don't pre-fetch resource lists unless you need their IDs — aggregate endpoints often return everything in one call.

Endpoint namespacing

Many resources live under sub-paths — never assume top-level. Common groups:

• /api/v1/auth/* — login, logout, refresh, lookup-token
• /api/v1/networking/* — data-networks, routes, NAT, BGP, interfaces
• /api/v1/ran/* — radios, events
• /api/v1/logs/audit — audit logs
• /api/v1/operator/* — operator code, ID, home-network-keys, NAS security
• /api/v1/cluster/* — HA cluster management
• /api/v1/subscribers, /api/v1/profiles, /api/v1/policies, /api/v1/slices — top-level
• /api/v1/subscriber-usage, /api/v1/flow-reports — top-level

When in doubt, fetch the OpenAPI spec.

Response envelope

• Success: {"result": <payload>}
• Error: {"error": "message"}
• Mutation success (201/200): {"result": {"message": "..."}}

Pagination

List endpoints accept page (default 1) and per_page (default 25, max 100). Responses include items, page, per_page, total_count. Use per_page=100 when iterating to minimize round-trips.

Mutations and destructive operations

For any POST, PUT, PATCH, or DELETE:

1. Confirm with the user first. State what will change and ask for explicit approval. Do not chain destructive calls without re-confirming.
2. List or GET before delete. Confirm the target exists and show its current state to the user.
3. Verify prerequisites before create. Resources follow a strict dependency chain (data network → slice → profile → policy → subscriber). See references/data-model.md.
4. One change at a time. Don't bundle unrelated mutations. Re-read state after each one to confirm it landed.

When to load references

External documentation

For non-API guidance, fetch from https://docs.ellanetworks.com/ — for example how_to/ai_agents/, how_to/backup_and_restore/, or reference/.