> ## Documentation Index > Fetch the complete documentation index at: https://docs.mka1.com/llms.txt > Use this file to discover all available pages before exploring further. # Create a run > Rent one GPU node from one provider, start one container from a public image, and run the supplied bash script in it. Exactly one of the two budget forms is required: hours bounds time exactly and cost approximately, max_usd the reverse. ## OpenAPI ````yaml https://apigw.mka1.com/speakeasy.json post /api/v1/serving/runs openapi: 3.1.1 info: title: MKA1 API version: 1.1.0 description: >- The MKA1 API is a RESTful API that provides access to the MKA1 platform. Learn how to get started with the API and the TypeScript SDK [here](https://mka1.apidocumentation.com/guides/getting-started). license: name: Proprietary servers: - url: https://apigw.mka1.com description: MKA1 API Gateway - url: / description: Relative server URL (configurable via SDK constructor) security: [] tags: - name: Resource Authorization description: >- Manage permissions for LLM resources. Create resources, grant/revoke permissions, and delete resources. Only resource owners can grant, revoke, or delete permissions. x-displayName: Resource Authorization - name: Embeddings description: >- Text embedding API endpoints for generating vector representations of text. Create semantic embeddings for search, clustering, and similarity matching using various embedding models. x-displayName: Embeddings - name: Feedback description: >- User feedback API for rating and commenting on chat completions. Collect thumbs up/down ratings and detailed feedback to improve model responses and track user satisfaction. x-displayName: Feedback - name: Images description: >- Image generation API endpoints for creating images from text descriptions. Generate images with control over size, quality, and style. x-displayName: Images - name: MCP Vault description: >- MCP vault API for storing user-owned MCP server configurations and encrypted credentials. Agents reference vault IDs so secrets are resolved only at tool execution time. x-displayName: MCP Vault - name: Speech description: >- Speech API endpoints for audio processing. Convert text to natural-sounding speech (TTS) or transcribe speech to text (STT) in different languages. x-displayName: Speech - name: Usage description: >- Usage tracking and analytics API for monitoring token consumption, request counts, and cost analysis. View detailed statistics per user, model, and time period. x-displayName: Usage - name: Extract description: >- Structured data extraction API for extracting information from files. Define JSON schemas to extract structured data from images, PDFs, and documents. Supports reusable schema templates. x-displayName: Extract - name: Text Classification description: >- Text classification API for categorizing text into predefined labels. Use AI models to classify text content for sentiment analysis, topic categorization, and content moderation. x-displayName: Text Classification - name: Responses description: >- Agent-powered responses API for creating AI agents with autonomous tool usage. Build conversational assistants that can use web search, file operations, image generation, code execution, computer use simulation, and MCP integrations. Supports background processing, streaming, and real-time status tracking. x-displayName: Responses - name: Files description: >- File management API for uploading, storing, and managing files with automatic expiration and S3 integration. Upload files that can be used with Assistants, Vector Stores, and other features. Files are stored in S3 with metadata tracked in PostgreSQL. Supports automatic cleanup of expired files. x-displayName: Files - name: Vector Stores description: >- Vector store API for storing and searching documents using embeddings. Create vector stores, upload files with automatic chunking and embedding generation, and perform semantic search. Files are processed asynchronously using Temporal workflows for durability. Supports automatic cleanup of expired stores and LanceDB for efficient vector storage. x-displayName: Vector Stores - name: Conversations description: >- Conversation management API for storing and retrieving conversation state across Response API calls. Create conversations, add items (user messages, assistant messages, system messages), and maintain conversation history. Supports metadata tracking and multi-turn dialogue state management. x-displayName: Conversations - name: Guardrails description: >- AI safety guardrails API for configuring content moderation and security policies. Set up ban word lists, prompt injection detection, and system prompt leakage prevention. Guardrails apply to all requests from an account and can be tested before deployment. x-displayName: Guardrails - name: Models description: >- Model listing API for discovering available models. Returns model IDs, ownership, and metadata for all registered models in the gateway. x-displayName: Models - name: Skills description: >- Skills API for managing versioned bundles of instructions and files following the Agent Skills standard. Create, version, and download reusable skill packages that include SKILL.md manifests for agent environments. x-displayName: Skills - name: Chat Completions description: >- **Deprecated: Use the Responses API (`/api/v1/llm/responses`) instead.** Chat completion endpoints with support for streaming, tool calls, and multiple providers. x-deprecated: true x-displayName: Chat Completions - name: Batches x-displayName: Batches - name: Evals x-displayName: Evals - name: Fine-Tuning x-displayName: Fine-Tuning - name: Memory Stores x-displayName: Memory Stores - name: Prompts x-displayName: Prompts - name: API Key x-displayName: API Key - name: Organization x-displayName: Organization - name: Cluster Admin x-displayName: Cluster Admin - name: Sessions description: Create, inspect, access, and terminate sandbox sessions. x-displayName: Sessions - name: Browser description: >- Connect to browser sessions through the gateway port proxy. Browser sessions expose a Chrome DevTools Protocol endpoint on port 9222. x-displayName: Browser - name: Execution description: Run shell commands and code inside an existing sandbox session. x-displayName: Execution - name: Workspace description: >- Inspect the workspace manifest, transfer files or archives, and download generated artifacts. x-displayName: Workspace - name: Sandbox Usage description: >- Aggregate sandbox usage statistics across sessions, execution, and workspace operations. x-displayName: Sandbox Usage - name: Sandbox Pricing description: >- Cluster-admin management of the sandbox compute rate card used for budgeted spend. x-displayName: Sandbox Pricing - name: Agents description: Create and manage reusable agent definitions. x-displayName: Agents - name: Agent Versions description: Inspect an agent's configuration history and roll back to a prior version. x-displayName: Agent Versions - name: Agent Runs description: Execute saved agents and inspect persisted run results. x-displayName: Agent Runs - name: Agent Connectors description: >- Connect saved agents to external messaging channels such as Telegram, including text, photo, and supported document exchange. x-displayName: Agent Connectors - name: Agent Schedules description: Create and manage scheduled or recurring saved agent runs. x-displayName: Agent Schedules - name: schema-4_other x-displayName: other - name: Budgets x-displayName: Budgets - name: Settings x-displayName: Settings - description: Long-lived inference servers. name: Deployments x-displayName: Deployments - description: Submit, monitor, and cancel fine-tune jobs. name: Fine-Tune Jobs x-displayName: Fine-Tune Jobs - description: Custom container images for deployments and jobs. name: Container Images x-displayName: Container Images - description: >- First-class training datasets: registered references or uploaded, committed objects. name: Datasets x-displayName: Datasets - description: Persistent storage for weights and checkpoints. name: Volumes x-displayName: Volumes - description: Credentials injected into your workloads. name: Secrets x-displayName: Secrets - description: Available accelerator types (GPU, NPU, TPU). name: Accelerators x-displayName: Accelerators - description: Provider-anonymous, pre-submission capacity eligibility previews. name: Eligibility x-displayName: Eligibility - description: Cluster-admin organization provisioning for serving. name: Serving Tenants x-displayName: Serving Tenants - description: >- Cluster-admin managed catalog of serving runtime recipes. Registration stores a digest-pinned manifest; only enabled recipes are deployable. name: Serving Recipes x-displayName: Serving Recipes - description: Computed cost and spend aggregates. name: Spend x-displayName: Spend paths: /api/v1/serving/runs: post: tags: - Fine-Tune Jobs summary: Create a run description: >- Rent one GPU node from one provider, start one container from a public image, and run the supplied bash script in it. Exactly one of the two budget forms is required: hours bounds time exactly and cost approximately, max_usd the reverse. operationId: createServingRun parameters: - name: X-On-Behalf-Of in: header required: false schema: type: string description: Optional external end-user identifier forwarded by the API gateway. requestBody: content: application/json: schema: $ref: '#/components/schemas/RunCreate' required: true responses: '201': content: application/json: schema: $ref: '#/components/schemas/Run' description: The run, accepted and submitted for acquisition. '401': content: application/json: schema: $ref: '#/components/schemas/ServingError' description: Missing or invalid bearer token. '403': content: application/json: schema: $ref: '#/components/schemas/ServingError' description: Serving is not enabled for your tenant. '404': content: application/json: schema: $ref: '#/components/schemas/ServingError' description: Referenced tenant resource not found. '409': content: application/json: schema: $ref: '#/components/schemas/ServingError' description: A live deployment with this organization-scoped name already exists. '422': content: application/json: schema: $ref: '#/components/schemas/ServingHTTPValidationError' description: Validation Error '500': content: application/json: schema: $ref: '#/components/schemas/ServingError' description: Unexpected server error. '503': content: application/json: schema: $ref: '#/components/schemas/ServingError' description: Deployment persistence or a required dependency is unavailable. security: - bearerAuth: [] x-codeSamples: - lang: python label: Python (SDK) source: |- from meetkai_mka1 import SDK with SDK( bearer_auth="", ) as sdk: res = sdk.fine_tune_jobs.create_serving_run(accelerator="", deadline={}, disk_gb=369181, gpu_count=444268, image="https://loremflickr.com/2112/3769?lock=8429739459860585", kind="deployment", startup_script_b64="") # Handle response print(res) - lang: typescript label: Typescript (SDK) source: |- import { SDK } from "@meetkai/mka1"; const sdk = new SDK({ bearerAuth: "", }); async function run() { const result = await sdk.fineTuneJobs.createServingRun({ runCreate: { accelerator: "", deadline: {}, diskGb: 369181, gpuCount: 444268, image: "https://loremflickr.com/2112/3769?lock=8429739459860585", kind: "deployment", startupScriptB64: "", }, }); console.log(result); } run(); - lang: csharp label: CSharp (SDK) source: >- using MeetKai.MKA1; using MeetKai.MKA1.Types.Components; var sdk = new SDK(bearerAuth: ""); var res = await sdk.FineTuneJobs.CreateServingRunAsync(body: new RunCreate() { Accelerator = "", Deadline = new RunDeadline() {}, DiskGb = 369181, GpuCount = 444268, Image = "https://loremflickr.com/2112/3769?lock=8429739459860585", Kind = RunCreateKind.Deployment, StartupScriptB64 = "", }); // handle response components: schemas: RunCreate: properties: accelerator: description: Curated accelerator name. GET /accelerators lists what is available. title: Accelerator type: string deadline: $ref: '#/components/schemas/RunDeadline' disk_gb: description: >- Required, no default. Must fit the image plus whatever the script downloads; too small and the pod dies mid-download. minimum: 1 title: Disk Gb type: integer env: additionalProperties: type: string description: >- Environment for the workload. Keys in the MKA1_ namespace are reserved. title: Env type: object gpu_count: description: >- Accelerators on one node. V1 is single-node: multi-node clusters are not offered by the providers we support. maximum: 8 minimum: 1 title: Gpu Count type: integer image: description: >- Any public image. It must contain bash, and on providers where we install sshd it must be Debian-family with apt. title: Image type: string interconnect: description: >- Optional filter on intra-node topology, e.g. nvlink. Tensor-parallel work is materially slower over PCIe. title: Interconnect type: string kind: description: >- What the run is for. A fine-tune ends on its own and its exit code is the outcome; a deployment serves traffic and runs until it is deleted. They are one resource because they are one pod, one provider and one script — the kind selects terminal semantics, not a different machine. enum: - finetune - deployment title: Kind type: string name: title: Name type: string ports: description: >- Container ports to expose, e.g. "8000/http". Meaningful for a deployment, which serves traffic; a fine-tune has nothing to expose. items: type: string title: Ports type: array providers: description: >- Ordered preference, tried in turn. Rejected while provider anonymity is enabled, because the list is itself a provider-identity leak. items: type: string title: Providers type: array secrets: description: >- Stored secret ids resolved into the environment immediately before hand-off. items: type: string title: Secrets type: array ssh_public_key: description: >- Your own public key. We never generate or hold private key material. Omit and the run has no SSH. title: Ssh Public Key type: string startup_script_b64: description: >- Base64 bash. Run as a child process of our run script; its exit code is the run's outcome. Never interpreted. title: Startup Script B64 type: string required: - kind - accelerator - gpu_count - image - startup_script_b64 - disk_gb - deadline title: RunCreate type: object Run: properties: accelerator: title: Accelerator type: string created_at: type: string format: date-time title: Created At deadline_at: format: date-time title: Deadline At type: string endpoint: description: >- Where the workload serves, once it is running. Only a deployment declares ports, so only a deployment reports an endpoint. title: Endpoint type: string exit_code: description: >- The workload's exit status. Provider status is not trusted for this: one provider reports the VM rather than the container. title: Exit Code type: integer gpu_count: title: Gpu Count type: integer id: title: Id type: string image: title: Image type: string kind: description: >- What the run is for. A fine-tune ends on its own and its exit code is the outcome; a deployment serves traffic and runs until it is deleted. They are one resource because they are one pod, one provider and one script — the kind selects terminal semantics, not a different machine. enum: - finetune - deployment title: Kind type: string name: title: Name type: string price_usd_hr: title: Price Usd Hr type: number provider: title: Provider type: string reason: description: >- Why the run is in this state, e.g. ssh_setup_failed or budget_exhausted. title: Reason type: string ssh: $ref: '#/components/schemas/RunSSH' started_at: format: date-time title: Started At type: string state: enum: - requested - provisioning - running - succeeded - failed - terminated title: State type: string terminal_at: format: date-time title: Terminal At type: string required: - id - kind - state - created_at title: Run type: object ServingError: description: Standard error envelope. properties: error: $ref: '#/components/schemas/ErrorDetail' required: - error title: ServingError type: object ServingHTTPValidationError: properties: detail: items: $ref: '#/components/schemas/ServingValidationError' title: Detail type: array title: ServingHTTPValidationError type: object RunDeadline: description: >- Exactly one of hours or max_usd. Cost is rate x time and the rate is fixed at acquisition, so they are interchangeable — but they are not the same number. A max_usd deadline resolves only once acquisition fixes the rate. properties: hours: exclusiveMinimum: 0 title: Hours type: number max_usd: exclusiveMinimum: 0 title: Max Usd type: number required: [] title: RunDeadline type: object RunSSH: description: >- Connection details. The login user is provider dependent and is read from the provider, never assumed. properties: command: description: >- Ready-to-use SSH invocation. Assembled here because the login user is provider dependent — it is read from the provider, never assumed — and rebuilding it by hand is where callers get it wrong. title: Command type: string host: title: Host type: string port: title: Port type: integer user: title: User type: string required: - host - port - user - command title: RunSSH type: object ErrorDetail: description: The body of an error response. properties: code: examples: - tenant_not_provisioned title: Code type: string message: type: string title: Message type: examples: - authorization_error title: Type type: string required: - code - message - type title: ErrorDetail type: object ServingValidationError: properties: loc: description: >- JSON path to the offending input: the first segment names where it lives ("body", "query", "path", or "request" when no narrower location is known), followed by property names / array indexes, e.g. ["body", "scaling", "min_containers"]. items: anyOf: - type: string - type: integer title: Location type: array msg: type: string title: Message type: type: string title: Error Type required: - loc - msg - type title: ServingValidationError type: object securitySchemes: bearerAuth: type: http scheme: bearer bearerFormat: API Key description: >- Gateway auth: send `Authorization: Bearer `. For multi-user server-side integrations, you can also send `X-On-Behalf-Of: `. ````