POST/api/models/{slug}/providers

Add a deployment to a model

On a public model the new deployment is your organization's private route to it; on your own model it extends the model. Always customer_managed. Idempotent on the deployment identity: a replay answers 200. A deployment with a base_url needs a verified payment method and counts toward 10 new endpoints per hour. Add the returned id to the waterfall to route through it.

認証: AuthorizationヘッダーにBearer形式のAPIキー

パスパラメータ

  • slugstring必須

    Model slug.

リクエストボディ

  • providerenum必須
    "openai""anthropic""gemini""azure_openai""openrouter""bedrock""local""fireworks""modal""experiential_cloud"
  • provider_model_idstring必須
  • base_urlstring | null

    Required for local and modal; forbidden for fixed-origin providers. An explicit http(s) endpoint with optional port and path, no userinfo, query or fragment.

  • regionstring | null
  • api_versionstring | null
  • provider_connection_idstring | null

    Pin to one of your API-key BYOK connections of the same provider; null resolves by provider at request time.

  • input_nano_usd_per_millioninteger | null

    Integer nano-USD per million tokens; null = unknown.

  • cached_input_nano_usd_per_millioninteger | null
  • cache_write_input_nano_usd_per_millioninteger | null
  • cache_write_1h_input_nano_usd_per_millioninteger | null
  • long_context_cache_write_input_nano_usd_per_millioninteger | null
  • long_context_cache_write_1h_input_nano_usd_per_millioninteger | null
  • output_nano_usd_per_millioninteger | null
  • reasoning_nano_usd_per_millioninteger | null
  • pricing_sourcestring | null

    Where the rates come from (for example self-reported). provider-free-tier is reserved and refused.

  • capabilitiesobject

    Wire capabilities. Seeded from the model's supported_params; explicit keys win.

    デフォルト: {}
  • uptime_30dnumber | null
  • throughput_tpsnumber | null
  • latency_p50_msnumber | null
  • stats_sourceenum
    "openrouter""observed""estimate"null
  • endpoint_api_keystring | null

    local only. Sent to your endpoint as Authorization: Bearer <key>. Stored in a vault, never returned.

  • org_idstring | null

    Owning organization. Implied by a key.

レスポンス

  • 201

    Created. A replay answers 200 with the same row.

    レスポンス
    • idstring

      Deployment id, or byok:{connection_id} for a lane served by your own provider connection. This is the token a waterfall write uses.

    • model_idstring
    • providerstring

      Provider id as shown to you; platform lanes run on our own capacity show experiential_cloud.

    • provider_model_idstring

      The provider's wire id.

    • base_urlstring | null

      Endpoint of a customer-run (local, modal) deployment.

    • regionstring | null
    • api_versionstring | null
    • owning_org_idstring | null

      Your organization for a private lane; null for a public lane.

    • provider_connection_idstring | null

      The pinned BYOK connection, if any.

    • connection_aliasstring | null

      On BYOK lanes only: the connection's account alias, which tells two same-provider accounts apart.

    • route_idstring | null

      Opaque per-organization handle for selecting exactly this lane on a request. Null without an organization context or when the lane cannot serve your traffic.

    • provider_statusobject

      The provider's own public status report for this lane. Advisory only, never routing input.

      provider_status
      • stateenum
        "incident_reported""no_incident_reported""stale_unavailable""no_verified_feed"
      • official_urlstring | null
      • checked_atstring | null
      • scopestring
      • limitationstring
      • incidentsarray<object>
      • advisoryboolean
    • provider_logo_urlstring

      Registry provider logo, when one exists. Omitted otherwise.

    • billing_sourceenum

      host_managed bills your credits; customer_managed bills your own provider or endpoint.

      "host_managed""customer_managed"
    • input_nano_usd_per_millioninteger | null
    • cached_input_nano_usd_per_millioninteger | null
    • cache_write_input_nano_usd_per_millioninteger | null
    • cache_write_1h_input_nano_usd_per_millioninteger | null
    • long_context_cache_write_input_nano_usd_per_millioninteger | null
    • long_context_cache_write_1h_input_nano_usd_per_millioninteger | null
    • cache_write_12h_input_nano_usd_per_millioninteger | null

      Null on a lane supporting 12-hour cache means 2.5 times the input rate.

    • output_nano_usd_per_millioninteger | null
    • reasoning_nano_usd_per_millioninteger | null
    • image_output_nano_usd_per_millioninteger | null

      Generated-image tokens rate; null bills them as output.

    • long_context_threshold_tokensinteger | null

      Once reported input tokens reach this, the long_context_* rates price the whole request. Null means flat pricing.

    • long_context_input_nano_usd_per_millioninteger | null
    • long_context_cached_input_nano_usd_per_millioninteger | null
    • long_context_output_nano_usd_per_millioninteger | null
    • long_context_reasoning_nano_usd_per_millioninteger | null
    • unit_pricesobject

      A per-unit lane's only meter.

      unit_prices
      • kindenum
        "character""audio_second""video_second""image"
      • ratesobject

        Integer nano-USD per unit by variant ("" is the flat rate).

    • off_peak_pricingobject

      Off-peak price card with its weekly UTC peak windows (prices, peak_windows, off_peak_dates). The headline rates are the peak price. Omitted when absent.

    • pricing_sourcestring | null
    • pricing_effective_atstring | null
    • capabilitiesobject

      Wire capabilities the gateway enforces (for example supports_tools, supported_reasoning_efforts).

    • uptime_30dnumber | null

      Percent.

    • uptime_recentnumber | null

      Completed share (percent) of the route's most recent requests; null under 50 requests.

    • uptime_recent_requestsinteger | null
    • throughput_tpsnumber | null

      Observed decode tokens per second.

    • latency_p50_msnumber | null

      Observed median time to first token.

    • request_latency_p50_msnumber | null

      Observed median end-to-end latency of calls with no first token.

    • stats_sourceenum
      "openrouter""observed""estimate"null
    • statusstring
    • routableboolean | null

      Whether the gateway's active plan routes this platform lane. Null when not applicable.

    • requires_paymentboolean

      Paid-only lane (a completed top-up opens it).

    • requires_purchaseboolean

      Purchase-required lane (a real purchase, not the $1 verification, opens it).

    • zdr_onlyboolean

      Serves only requests routed under zero data retention.

    • hidden_fallbackboolean
    • available_to_orgboolean | null

      Whether your organization may use this closed lane. Null without an organization context or for an open lane.

    • zero_data_retentionboolean | null

      Detail route only. Null means no posture coverage, which a require-ZDR policy treats as non-compliant.

    • zero_data_retention_on_requestboolean | null

      ZDR only when dispatched under the gateway's per-request ZDR constraint.

    • processing_region_setsarray<array<string>>
    • plan_discountsobject | null
      いずれか

      object

      • freenumber
      • pronumber
      • maxnumber
      • ultranumber
    • service_tier_ratesobject | null

      Per-tier rate cards (flex, priority) in nano-USD per million tokens when the lane prices processing tiers.

    • ladder_rankinteger | null

      Display-only position among the model's public platform lanes.

    • ladder_lane_countinteger | null
    • ladder_scorenumber | null
    • ladder_factorsobject | null
    • endpoint_credential_last4string | null

      Last four characters of a stored endpoint API key.

    • created_atstring
    • updated_atstring
  • 401

    Missing, unknown or revoked API key.

    レスポンス
    • errorstring必須

      Human-readable message.

    • codestring

      Stable machine reason, when the route defines one (for example pro_required).

    • actionstring

      Stable remediation action, when the route defines one.

    • detailsobject

      Customer-safe string facts the caller needs to act (for example a suggested identifier).

  • 403

    The organization is under review (org_under_review), or a base_url deployment needs a verified payment method (endpoint_requires_verified_payment).

    レスポンス
    • errorstring必須

      Human-readable message.

    • codestring

      Stable machine reason, when the route defines one (for example pro_required).

    • actionstring

      Stable remediation action, when the route defines one.

    • detailsobject

      Customer-safe string facts the caller needs to act (for example a suggested identifier).

  • 404

    The model, the organization, or the pinned provider_connection_id is not visible to you.

    レスポンス
    • errorstring必須

      Human-readable message.

    • codestring

      Stable machine reason, when the route defines one (for example pro_required).

    • actionstring

      Stable remediation action, when the route defines one.

    • detailsobject

      Customer-safe string facts the caller needs to act (for example a suggested identifier).

  • 409

    The deployment identity exists with different attributes.

    レスポンス
    • errorstring必須

      Human-readable message.

    • codestring

      Stable machine reason, when the route defines one (for example pro_required).

    • actionstring

      Stable remediation action, when the route defines one.

    • detailsobject

      Customer-safe string facts the caller needs to act (for example a suggested identifier).

  • 422

    Body validation failed (for example endpoint_api_key on a non-local provider, or a base_url with a query string), or a schema rule such as pricing_source_reserved.

  • 429

    The organization created 10 custom endpoints in the last hour (endpoint_creation_rate_limited).

    レスポンス
    • errorstring必須

      Human-readable message.

    • codestring

      Stable machine reason, when the route defines one (for example pro_required).

    • actionstring

      Stable remediation action, when the route defines one.

    • detailsobject

      Customer-safe string facts the caller needs to act (for example a suggested identifier).

リクエスト

curl -X POST https://api.experientiallabs.ai/api/models/my-local-model/providers \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "provider": "local",
  "provider_model_id": "my-model-fp8",
  "base_url": "https://gpu2.acme.dev:8000/v1",
  "endpoint_api_key": "sk-acme-endpoint-9b3e5d1a7c",
  "pricing_source": "self-reported"
}'

レスポンス

{
  "id": "7a3f1c28-9e4d-4b60-a5d2-3c8e1f0b7d46",
  "model_id": "0b8f3d2a-5c61-4e7f-9a12-7d4e6b1c9f30",
  "provider": "local",
  "provider_model_id": "my-model-fp8",
  "base_url": "https://gpu2.acme.dev:8000/v1",
  "region": null,
  "api_version": null,
  "owning_org_id": "3f6c2a1e-8b7d-4c2a-9e51-6a0d4b2f7c10",
  "provider_connection_id": null,
  "billing_source": "customer_managed",
  "input_nano_usd_per_million": null,
  "cached_input_nano_usd_per_million": null,
  "cache_write_input_nano_usd_per_million": null,
  "cache_write_1h_input_nano_usd_per_million": null,
  "long_context_cache_write_input_nano_usd_per_million": null,
  "long_context_cache_write_1h_input_nano_usd_per_million": null,
  "output_nano_usd_per_million": null,
  "reasoning_nano_usd_per_million": null,
  "long_context_threshold_tokens": null,
  "long_context_input_nano_usd_per_million": null,
  "long_context_cached_input_nano_usd_per_million": null,
  "long_context_output_nano_usd_per_million": null,
  "long_context_reasoning_nano_usd_per_million": null,
  "pricing_source": "self-reported",
  "pricing_effective_at": null,
  "capabilities": {},
  "uptime_30d": null,
  "throughput_tps": null,
  "latency_p50_ms": null,
  "stats_source": null,
  "status": "active",
  "endpoint_credential_last4": "7c1a",
  "created_at": "2026-11-19T15:10:42.330118+00:00",
  "updated_at": "2026-11-19T15:10:42.330118+00:00"
}

https://api.experientiallabs.ai/api/models/my-local-model/providers