> ## Documentation Index > Fetch the complete documentation index at: https://docs.hyperline.co/llms.txt > Use this file to discover all available pages before exploring further. # Ingest AI calls > Ingest a batch of AI records (limited to max 2000 per request) — model calls and non-LLM cost items, told apart by `record_kind` (omitted means `llm_call`). Records are processed asynchronously; a record with the same id is deduplicated on storage. ## OpenAPI ````yaml post /v1/ai/calls openapi: 3.1.0 info: title: Hyperline Ingest API version: 0.0.0 servers: - url: https://ingest.hyperline.co - url: https://sandbox.ingest.hyperline.co security: [] paths: /v1/ai/calls: post: tags: - AI summary: Ingest AI calls description: >- Ingest a batch of AI records (limited to max 2000 per request) — model calls and non-LLM cost items, told apart by `record_kind` (omitted means `llm_call`). Records are processed asynchronously; a record with the same id is deduplicated on storage. operationId: ingestAiCalls requestBody: description: AI calls payload. content: application/json: schema: type: array items: type: object properties: id: type: string minLength: 1 maxLength: 128 description: >- Caller-generated record id. A record with the same id is deduplicated on storage. example: aic_01JT4E2Q4N8N0F0X4S2M3R7YQK run_id: type: string minLength: 1 maxLength: 128 description: >- The run this record belongs to. Every record captured inside one `hyperline.run()` shares it. example: run_01JT4E2Q4N8N0F0X4S2M3R7YQL parent_run_id: type: - string - 'null' minLength: 1 maxLength: 128 description: >- The enclosing run, when this one was opened inside another. Null at the top of a run tree. example: null customer_id: type: string minLength: 1 maxLength: 255 description: >- The customer this usage is attributed to — the external id your application knows them by, or a Hyperline `cus_…` id. example: cus_xyz789 agent: type: string minLength: 1 maxLength: 128 description: Stable agent identity, e.g. the agent's name or type. example: invoice-copilot action: type: - string - 'null' minLength: 1 maxLength: 128 description: >- What the run does, e.g. `deep_research`. Null when unnamed. example: monitoring_analysis record_kind: type: string enum: - llm_call - cost default: llm_call description: >- What the record is. `llm_call` is a model invocation and the token counters mean something; `cost` is a non-LLM cost item whose quantity rides in `usage_extra`. Absent means `llm_call`, so an older SDK keeps ingesting unchanged. example: llm_call provider: type: string minLength: 1 maxLength: 64 description: The model's provider, or the vendor on a `cost` record. example: anthropic model: type: string minLength: 1 maxLength: 128 description: >- The model invoked, or the item purchased on a `cost` record. example: claude-sonnet-5 input_tokens: type: integer minimum: 0 maximum: 1000000000 description: Uncached input tokens. example: 12400 output_tokens: type: integer minimum: 0 maximum: 1000000000 description: Output tokens. example: 1300 cached_input_tokens: type: integer minimum: 0 maximum: 1000000000 description: Input tokens served from the provider's prompt cache. example: 9000 cache_write_tokens: type: integer minimum: 0 maximum: 1000000000 description: Input tokens written into the provider's prompt cache. example: 0 reasoning_tokens: type: integer minimum: 0 maximum: 1000000000 description: Reasoning tokens, on models that bill them separately. example: 0 status: type: string enum: - ok - error description: Whether the call succeeded. example: ok duration_ms: type: integer minimum: 0 description: Wall-clock duration of the call. example: 1840 started_at: type: string format: date-time description: When the call started, as an ISO instant. example: '2026-08-18T10:22:31.201Z' metadata: type: - object - 'null' additionalProperties: anyOf: - type: string maxLength: 1024 - type: number - type: boolean - type: 'null' - type: 'null' description: >- Free-form bag carried alongside the record. At most 50 keys, string values at most 1024 characters, 4096 bytes once serialized. example: conversation_id: cnv_123 usage_extra: type: - object - 'null' additionalProperties: type: number minimum: 0 maximum: 1000000000000 description: >- Cost dimensions beyond the token classes — provider extras (web searches, image generations) or the caller's own units (vector queries, gpu_seconds). Keys must match `[a-z0-9_.]{1,64}`, values must be finite and at most 1e12, at most 20 keys. example: web_search: 2 required: - id - run_id - parent_run_id - customer_id - agent - action - provider - model - input_tokens - output_tokens - cached_input_tokens - cache_write_tokens - reasoning_tokens - status - duration_ms - started_at - metadata description: >- One record captured by the Hyperline AI SDK — a model call (`record_kind: "llm_call"`, the default when the field is absent) or a non-LLM cost item (`record_kind: "cost"`, where `provider`/`model` carry the vendor/item and the quantity rides in `usage_extra`). example: id: aic_01JT4E2Q4N8N0F0X4S2M3R7YQK run_id: run_01JT4E2Q4N8N0F0X4S2M3R7YQL parent_run_id: null customer_id: cus_xyz789 agent: invoice-copilot action: monitoring_analysis record_kind: llm_call provider: anthropic model: claude-sonnet-5 input_tokens: 12400 output_tokens: 1300 cached_input_tokens: 9000 cache_write_tokens: 0 reasoning_tokens: 0 status: ok duration_ms: 1840 started_at: '2026-08-18T10:22:31.201Z' metadata: conversation_id: cnv_123 usage_extra: web_search: 2 minItems: 1 maxItems: 2000 responses: '202': description: Calls accepted. content: application/json: schema: type: object properties: calls_accepted: type: integer description: Number of calls accepted for asynchronous processing. example: 42 required: - calls_accepted security: - bearer: [] components: securitySchemes: bearer: type: http scheme: bearer bearerFormat: JWT ````