> ## Documentation Index > Fetch the complete documentation index at: https://docs.together.ai/llms.txt > Use this file to discover all available pages before exploring further. # Create an endpoint > Creates a stable, inference-addressable endpoint. Add one or more deployments and configure its traffic split before sending inference requests to the endpoint name. ## OpenAPI ````yaml POST /projects/{projectId}/endpoints openapi: 3.1.0 info: title: Together APIs description: The Together REST API. See https://docs.together.ai for more details. version: 2.0.0 termsOfService: https://www.together.ai/terms-of-service contact: name: Together Support url: https://www.together.ai/contact license: name: MIT url: https://github.com/togethercomputer/openapi/blob/main/LICENSE servers: - url: https://api.together.ai/v1 description: Default environment for APIs - url: https://api-inference.together.ai/v2 description: Optimized environment for inference security: - bearerAuth: [] paths: /projects/{projectId}/endpoints: post: tags: - EndpointService summary: Create an endpoint description: >- Creates a stable, inference-addressable endpoint. Add one or more deployments and configure its traffic split before sending inference requests to the endpoint name. operationId: EndpointService_CreateEndpoint parameters: - name: projectId in: path required: true schema: description: Project identifier. type: string requestBody: content: application/json: schema: $ref: '#/components/schemas/DE.CreateEndpointRequest' required: true responses: '200': description: OK content: application/json: schema: $ref: '#/components/schemas/DE.Endpoint' default: description: Default error response content: application/json: schema: $ref: '#/components/schemas/ErrorData' servers: - url: https://api.together.ai/v2 components: schemas: DE.CreateEndpointRequest: type: object description: Request parameters for creating an endpoint. required: - name properties: name: type: string description: Inference-addressable endpoint name to create. visibility: default: VISIBILITY_PRIVATE enum: - VISIBILITY_PRIVATE - VISIBILITY_INTERNAL type: string description: >- Who can discover the endpoint. `VISIBILITY_PRIVATE` restricts it to the project; `VISIBILITY_INTERNAL` shares it with the organization. DE.Endpoint: description: >- Stable inference entry point that groups deployments and routes requests among them. type: object required: - name - id - projectId - createdAt - updatedAt - etag - deployments - trafficSplit - visibility - endpointType properties: id: readOnly: true description: Unique endpoint identifier. type: string projectId: readOnly: true type: string description: ID of the project that owns the endpoint. name: type: string description: >- Project-qualified endpoint name in the form `/`. Pass this value as `model` in inference requests. Create and update requests may use either a bare endpoint name or the qualified form; a supplied project slug must match the project in the request path. createdAt: readOnly: true type: string description: Timestamp when the endpoint was created. format: date-time updatedAt: type: string description: Output only. Timestamp when the endpoint was last updated. format: date-time etag: type: string description: |- Opaque version tag for optimistic concurrency control. Supply on update/delete to ensure consistent read-modify-write. If not set, the write overwrites based on current state. trafficSplit: type: array items: $ref: '#/components/schemas/DE.TrafficSplitEntry' description: >- Deployments eligible for live traffic and their capacity weights. An empty list leaves the endpoint unrouted. visibility: default: VISIBILITY_PRIVATE enum: - VISIBILITY_PRIVATE - VISIBILITY_INTERNAL type: string description: >- Who can discover the endpoint. `VISIBILITY_PRIVATE` restricts it to the project; `VISIBILITY_INTERNAL` shares it with the organization. endpointType: readOnly: true enum: - ENDPOINT_TYPE_DEDICATED - ENDPOINT_TYPE_SERVERLESS type: string description: Serving class of the endpoint. deployments: type: array readOnly: true items: $ref: '#/components/schemas/DE.DeploymentSummary' description: >- Lightweight summaries of deployments under this endpoint. Retrieve a deployment through the endpoint's deployment API for full details. activeRolloutId: readOnly: true type: string description: ID of the currently active rollout, or empty if none. ErrorData: type: object required: - error properties: error: type: object properties: message: type: string nullable: false type: type: string nullable: false param: type: string nullable: true default: null code: type: string nullable: true default: null required: - type - message DE.TrafficSplitEntry: type: object description: >- Capacity weight assigned to one deployment in an endpoint's live traffic split. required: - deploymentId - weight properties: deploymentId: type: string description: ID of a deployment under the endpoint that can receive live traffic. weight: type: number description: >- Non-negative, finite weight applied to each ready replica. A deployment's effective routing capacity is `weight * readyReplicas`, and requests are distributed in proportion to that capacity. Set to `0` to remove the deployment from the live traffic split. DE.DeploymentSummary: type: object description: Compact deployment status embedded in an endpoint response. required: - id - autoscaling - createdAt - hardware - model - modelId - name - state - trafficMode properties: id: readOnly: true type: string description: Deployment identifier. modelId: type: string description: Deprecated. Use `model`. Model identifier being served. hardware: readOnly: true type: string description: >- Hardware configuration selected by the deployment's config, such as its GPU type and count. trafficMode: enum: - TRAFFIC_MODE_LIVE - TRAFFIC_MODE_SHADOW type: string description: >- Whether the deployment serves client-visible responses or only mirrored shadow traffic. state: readOnly: true enum: - DEPLOYMENT_STATE_PROVISIONING - DEPLOYMENT_STATE_READY - DEPLOYMENT_STATE_SCALING - DEPLOYMENT_STATE_DEGRADED - DEPLOYMENT_STATE_FAILED - DEPLOYMENT_STATE_STOPPED - DEPLOYMENT_STATE_STOPPING type: string description: Current state of the deployment. name: readOnly: true type: string description: >- Inference-addressable name in the fully-qualified form "//". Pass it as the "model" field when calling the inference API to pin to this deployment. createdAt: readOnly: true type: string description: Timestamp when the deployment was created. format: date-time autoscaling: readOnly: true allOf: - $ref: '#/components/schemas/DE.AutoscalingResponse' description: Autoscaling configuration for the deployment. desiredReplicas: readOnly: true type: integer description: >- Number of replicas the autoscaler currently wants across all regions. readyReplicas: readOnly: true type: integer description: >- Number of replicas currently ready to serve requests across all regions. model: type: string description: >- Resource name of the served model in the form `projects/{projectId}/models/{modelId}/revisions/{revisionId}`. For public models, the model's owning project may differ from the deployment's project. estimatedEffectiveTrafficShare: readOnly: true type: number format: double description: >- Estimated fraction from 0 to 1 of endpoint traffic currently routed to this deployment. DE.AutoscalingResponse: description: Resolved autoscaling configuration returned for a deployment. allOf: - $ref: '#/components/schemas/DE.Autoscaling' - type: object required: - minReplicas - maxReplicas DE.Autoscaling: type: object properties: minReplicas: type: integer description: >- Minimum number of replicas. Omit on update to preserve the current value. Set both `minReplicas` and `maxReplicas` to `0` to stop the deployment. maxReplicas: type: integer description: >- Maximum number of replicas. Defaults to `minReplicas`; omitting it on update preserves the current value. scaleDownWindow: pattern: ^-?(?:0|[1-9][0-9]{0,11})(?:\.[0-9]{1,9})?s$ type: string description: >- Time a lower replica recommendation must remain stable before scaling down. Defaults to `5m`. scaleUpWindow: pattern: ^-?(?:0|[1-9][0-9]{0,11})(?:\.[0-9]{1,9})?s$ type: string description: Stabilization window before scaling up. scaleToZeroWindow: pattern: ^-?(?:0|[1-9][0-9]{0,11})(?:\.[0-9]{1,9})?s$ type: string description: >- Idle period after which the deployment automatically stops and releases its replicas. scalingMetrics: type: array items: $ref: '#/components/schemas/DE.ScalingMetric' description: >- Metrics and targets that drive replica recommendations. When omitted, the platform uses concurrent in-flight requests per replica. description: Autoscaling configuration for a deployment. DE.ScalingMetric: type: object properties: name: type: string description: >- Metric name, such as `gpu_utilization`, `ttft`, `inflight_requests`, `e2e_latency`, `throughput_per_replica`, or `decoding_speed`. type: enum: - METRIC_TARGET_TYPE_VALUE - METRIC_TARGET_TYPE_UTILIZATION - METRIC_TARGET_TYPE_AVERAGE_VALUE type: string description: >- Whether `target` is an absolute value, a utilization percentage, or a per-replica average. target: type: number description: >- Target interpreted according to `type`. Utilization uses a percentage from 0 to 100, value uses an absolute measurement, and average value uses a per-replica measurement. percentile: type: string description: >- Percentile to evaluate for latency-based metrics: `p50`, `p90`, `p95`, or `p99`. description: Metric and target used by the autoscaler to recommend a replica count. required: - name - type - target securitySchemes: bearerAuth: type: http scheme: bearer x-bearer-format: bearer x-default: default ````