> ## Documentation Index > Fetch the complete documentation index at: https://docs.omnia-voice.com/llms.txt > Use this file to discover all available pages before exploring further. # Create or repoint an alias > Point a stable name at a model — and repoint it whenever you like, with **no change to your code**. Your application keeps calling `support-bot`; the next request after a repoint already routes to the new model. This is what makes an eval result actionable: prove a cheaper model holds up, then repoint. **Upsert by name**, so CI can repoint idempotently without a lookup first. Owner/admin key required — repointing silently redirects live production traffic, so it's audited with the before/after. **You are always billed for the model that RAN.** An alias is a routing decision, never a pricing one. ## OpenAPI ````yaml /openapi.json put /v1/aliases openapi: 3.1.0 info: title: Omnia Management API description: >- The management API behind the improvement loop: capture and setup, request logs and datasets, grades (labels), judges (criteria), evals and deploy gates, fine-tuning and reinforcement learning, dedicated GPU endpoints, and model aliases and versions. Authenticated with a workspace API key (sk_sovereign_...). The inference API (chat, embeddings, rerank, responses) is OpenAI-compatible and documented separately. Responses are snake_case, list endpoints on the loop products use the {"object": "list", "data": [...]} envelope, and refusals use the same nested error shape the gateway emits: {"error": {"message", "type", "code"}}. Request bodies on the loop products (logs, labels, criteria, evals, datasets, aliases) are snake_case; the training and infrastructure products (fine-tuning, GRPO, environment tools, dedicated, model-version adoption) validate camelCase bodies, and each schema below says which it is. Endpoints that spend money require a key minted by a workspace owner or admin and return 403 otherwise. version: 1.0.0 servers: - url: https://gateway.omnia-voice.com description: Production - url: https://platform.omnia-voice.com/api description: Production (legacy alias — same API, older base URL) security: - bearerAuth: [] paths: /v1/aliases: put: tags: - Model aliases summary: Create or repoint an alias description: >- Point a stable name at a model — and repoint it whenever you like, with **no change to your code**. Your application keeps calling `support-bot`; the next request after a repoint already routes to the new model. This is what makes an eval result actionable: prove a cheaper model holds up, then repoint. **Upsert by name**, so CI can repoint idempotently without a lookup first. Owner/admin key required — repointing silently redirects live production traffic, so it's audited with the before/after. **You are always billed for the model that RAN.** An alias is a routing decision, never a pricing one. operationId: upsertAlias requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/UpsertModelAlias' responses: '200': description: The alias. content: application/json: schema: $ref: '#/components/schemas/ModelAlias' '400': $ref: '#/components/responses/BadRequest' '401': $ref: '#/components/responses/Unauthorized' '403': $ref: '#/components/responses/Forbidden' '412': description: >- Evidence refusal: require_evidence is on and no finished, passing, in-window comparison proves the destination model against the traffic it would take over. Same status the /gate endpoint uses, so one pipeline handles both. Pass override_reason (at least 10 characters) for an audited override. content: application/json: schema: $ref: '#/components/schemas/Error' components: schemas: UpsertModelAlias: type: object required: - name - target_model properties: name: type: string example: support-bot description: >- 3–64 chars (letters, numbers, dots, dashes, underscores). Scoped to your workspace — a name that isn't an alias is treated as a normal model id, so aliases never shadow the catalog. target_model: type: string example: Qwen/Qwen3-32B description: The model requests route to. Validated against the live catalog. canary_model: type: - string - 'null' description: >- Optional. A share of requests route here instead — for a gradual cutover. canary_percent: type: integer minimum: 0 maximum: 100 default: 0 description: >- 0 = no split. 100 = full cutover. The split is per REQUEST, not per user. description: type: - string - 'null' maxLength: 200 example: switched after eval ev_123 (90% win, -88% cost) gate_criterion_id: type: - string - 'null' description: >- Turn the canary split into a supervised rollout: the online scorer judges BOTH arms with this criterion and the gate decides on corrected-rate CI bounds. The criterion must have been aligned against your labels. gate_mode: type: string enum: - recommend - auto default: recommend description: >- recommend: verdicts surface on GET /v1/aliases (gate_verdict) and via alert notifications — you click the button. auto: the gate may repoint the alias itself (promote) or zero the canary (roll back), always audited with the numbers. Auto requires an ALIGNED criterion (TPR and TNR ≥ 90%): a judge may recommend with a weaker signal, but it doesn't get the keys. gate_min_samples: type: integer minimum: 10 maximum: 1000 default: 50 description: >- No verdict of any kind until BOTH arms have this many scored requests. gate_rollback_threshold: type: number minimum: 0 maximum: 1 default: 0.7 description: >- Roll back only when even the canary's OPTIMISTIC bound (CI upper) sits below this — confidently bad, not unluckily sampled. gate_window_hours: type: integer minimum: 1 maximum: 720 default: 168 description: Scores older than this don't count toward the verdict. require_evidence: type: boolean description: >- Omitted = leave the alias's existing policy alone (a plain repoint can't quietly switch the policy off); false on create. With it on, a PUT that routes traffic to a model it isn't already reaching must be backed by a finished comparison in this workspace, within the last 30 days, whose candidate win-rate CI lower bound clears 50% against the model traffic is leaving. Refusals are 412. override_reason: type: - string - 'null' description: >- The audited escape hatch for an evidence refusal: a written justification of at least 10 characters lets the repoint through and writes an audit event naming the actor, both models, and the reason. Omitted or blank is not an override. ModelAlias: type: object properties: id: type: string name: type: string example: support-bot description: The name you put in `model`. target_model: type: string description: The model this resolves to today. canary_model: type: - string - 'null' canary_percent: type: integer description: 0–100. Share of requests routed to canaryModel. description: type: - string - 'null' created_at: type: string format: date-time updated_at: type: string format: date-time gate_criterion_id: type: - string - 'null' gate_mode: type: string enum: - recommend - auto gate_verdict: type: - object - 'null' description: >- The last gate evaluation: {decision: promote|rollback|hold, reason, canary: {model, rate, ci, judged}, incumbent: {...}, acted}. gate_verdict_at: type: - string - 'null' format: date-time gate_min_samples: type: integer description: >- No gate verdict until both canary arms have this many scored samples in the window. gate_rollback_threshold: type: number description: >- Roll back only when even the canary's optimistic bound (CI upper) sits below this corrected pass rate. gate_window_hours: type: integer description: Scores older than this do not count toward the verdict. model_version_id: type: - string - 'null' description: >- The model version this alias currently serves, when it points at a trained version (null for raw catalog models). Repoint to an older version to roll back. require_evidence: type: boolean description: >- Evidence policy: with this on, a repoint that would send traffic to an unproven model is refused with 412 unless a passing, in-window comparison exists (or an audited override_reason is given). last_evidence_run_id: type: - string - 'null' description: >- The eval run that authorized the current routing. Null when the routing predates the policy or went through as an audited override. Error: type: object properties: error: type: object properties: message: type: string type: type: string description: >- invalid_request_error, insufficient_quota, rate_limit_error, or api_error. code: type: string description: >- Machine-stable cause, e.g. invalid_api_key, not_found, insufficient_permissions, precondition_failed. required: - message - type - code description: >- Every refusal — gateway and management API alike — uses this one envelope. responses: BadRequest: description: Malformed request or invalid field. content: application/json: schema: $ref: '#/components/schemas/Error' Unauthorized: description: Missing, malformed, or revoked API key. content: application/json: schema: $ref: '#/components/schemas/Error' example: error: message: Invalid API key type: invalid_request_error code: invalid_api_key Forbidden: description: The key lacks the required owner/admin permission. content: application/json: schema: $ref: '#/components/schemas/Error' securitySchemes: bearerAuth: type: http scheme: bearer description: >- Your workspace API key, e.g. `sk_sovereign_...`, sent as `Authorization: Bearer `. ````