Nexus Mode
Nexus mode turns Cortex into a console for a Nexus hub.
Switch to it from the status bar or with Ctrl+M.
Connecting
Section titled “Connecting”Accesos → Connections holds the workspace’s hub connections:
| Field | Notes |
|---|---|
| Name | Any label |
| Hub URL | Presets for local (http://localhost:8080) and the hosted environments |
| Tenant ID | Optional |
| Auth | API key (sent as X-API-Key; the key already belongs to one tenant) or Email + contraseña (a JWT session, which can see every tenant of that user) |
A connection is saved only after it works: an API key is checked against
GET /api/v1/tenants, and an email connection asks you to sign in. JWT sessions refresh
themselves on a 401; if the refresh fails, Cortex signs you out.
Credentials are not stored in the workspace. Cortex keeps them encrypted with the operating system’s keychain, one set per machine, so switching to a connection with a different key replaces the stored one.
The Nexus layout
Section titled “The Nexus layout”| Activity | Panel |
|---|---|
| Tenants | The tenants this connection can see |
| Agentes | The tenant’s agents with their status; click one to open it |
| Uso | Consumption, by month, agent and invocation |
| Catalog | The hub’s built-in tools and integrations |
| Admin | Tariffs and plans, with an operator token |
| Accesos | API keys and connections |
A tenant selector at the top of the sidebar sets the tenant every panel works on. The right panel is the run panel.
Agents
Section titled “Agents”Publishing
Section titled “Publishing”An agent reaches the hub from the Runtime-mode editor: Deploy → Nexus Cloud → <connection>
sends its YAML to POST /api/v1/agents and reports Publicado: <name> (v<n>). Every publish
is a new version.
Editing
Section titled “Editing”Opening an agent from Agentes shows its current spec as YAML, validated against the same agent schema. Save publishes the edited spec as a new version; the trash icon deletes the agent after confirmation.
Versions
Section titled “Versions”Versiones opens the agent’s history: number, date, author and checksum, with the current version marked.
- Ver shows a version’s spec read-only.
- Comparar picks two versions and shows a side-by-side diff of their specs.
- Revertir republishes an old version’s spec as a new current version. History is kept; nothing is deleted.
Running agents
Section titled “Running agents”The run panel lists the tenant’s agents and preselects the one open in the editor. Write a
query and press Enviar or Ctrl/Cmd+Enter.
- Streaming. Cortex opens the hub’s run stream (
/api/v1/agents/{name}/run/stream) and shows tool calls as they happen. Stop cancels the run; the hub records it as cancelled and charges nothing. Against a hub without the stream route, Cortex falls back to a blocking run, which cannot be stopped. - Sessions. Nueva sesión starts a new conversation.
- Transcripts are kept per tenant on this machine, up to 50 finished turns per agent, and come back when you return to the tenant.
Each answer shows its status, invocation id, tokens and credits, then one row per model:
provider, model, role, calls, credits and cached tokens, with a mark when the model is the
platform’s own. Proposed writes appear as in
the runtime console, including vía <sub-agent>.
Hub errors are shown in plain words:
| Hub answer | Cortex says |
|---|---|
402 | The tenant ran out of credits |
404 | Agent not found |
422 | The agent has no published version |
429 | Concurrency or rate limit exceeded |
502 | The runtime is unavailable |
504 | Timed out (120 s) |
Consumption
Section titled “Consumption”Uso reads the hub’s observability endpoints in three levels:
| Level | Shows |
|---|---|
| Tenant | For the chosen month (last 12): invocations, credits, tokens in and out; all-time totals; agents ranked by credits, deleted ones marked |
| Agent | Invocations and credits in the month; recent invocations with status and credits |
| Invocation | Tokens, credits, duration, error, usage per model (with cached tokens) and its sub-invocations |
Catalog
Section titled “Catalog”Catalog lists the hub’s built-in tools and integrations, with each integration’s actions
and which ones write. Copiar YAML copies a ready tool entry to paste into an agent: a
builtin entry, or an integration entry with its read-only actions selected. If every action
writes, it lists them all with a reminder to review confirm.
API keys
Section titled “API keys”Accesos → API Keys lists the tenant’s keys by label and hash. Create one with a label; the raw key is shown once, so copy it then. The trash icon revokes a key immediately.
Operator: tariffs and plans
Section titled “Operator: tariffs and plans”Admin is for a hub operator. It asks for the hub’s operator token, sent only as
X-Operator-Token and kept apart from tenant credentials.
| Section | What you can do |
|---|---|
| Tarifas vigentes | Read current tariffs: model, credits per 1k tokens in and out, own-model mark |
| + tarifa | Add a tariff: provider, model, effective-from instant, credits per 1k in and out, cost in USD per 1k in and out, infrastructure cost, own model. Tariffs are append-only: a correction is a new row |
| Planes | Read plans with their included credits and limits |
| + plan | Create a plan: minimum fee, included credits, extra-credit price, activation price, and limits on active agents, concurrency, rpm, credits per period, tokens per call and iterations. An empty limit means no limit; 0 means blocked |
| Asignar plan | Assign a plan to a tenant by id |
Amounts are entered in dollars and sent to the hub in micros. See Plans & Billing for what each field does.