Adapters

Built-in adapters

apiexec ships four adapters, all registered at process startup via static initialisers. Pass the adapter name as the first argument to stream_create (or the equivalent in your binding). No explicit registration call is needed.

generic_rest

Works with any JSON REST endpoint that returns a next-page cursor. The adapter extracts a batch of records on each page, advances the cursor, and signals exhaustion when the cursor field is null.

Config keys

Example

{
  "base_url": "https://api.example.com/v1/events",
  "auth_header": "Bearer sk-abc123",
  "data_field": "events",
  "next_token_field": "next_cursor",
  "page_size": 200
}

StreamHandle* s = stream_create("generic_rest", config_json, policy_json);

datadog_metrics

Queries the Datadog Metrics API v1. Splits the time range into windows and streams one window per batch. The window size adapts automatically: it shrinks on 429 responses and grows when fetches succeed, converging toward max_window_ms.

Config keys

Example

{
  "api_key": "YOUR_DD_API_KEY",
  "app_key": "YOUR_DD_APP_KEY",
  "query": "avg:system.cpu.user{host:web-01}",
  "window_ms": 1800000
}

StreamHandle* s = stream_create("datadog_metrics", config_json, NULL);

openai

Sends one prompt per page through the OpenAI Chat Completions API. Token usage from each response is reported via response_cost() so that CostAwarePolicy can enforce a budget.

Config keys

Example

{
  "api_key": "sk-...",
  "model": "gpt-4o-mini",
  "prompts": ["Summarise the following text:", "Classify the sentiment:"],
  "max_tokens": 512,
  "temperature": 0.7
}

anthropic

Sends one prompt per page through the Anthropic Messages API. Token usage is tracked the same way as the openai adapter.

Config keys

Example

{
  "api_key": "sk-ant-...",
  "model": "claude-haiku-4-5-20251001",
  "prompts": ["Summarise: ...", "Extract entities from: ..."],
  "max_tokens": 256
}

Custom adapters

To add your own adapter, implement the VendorAdapter<T> interface and register it with StreamRegistry. See Architecture - Adapter Interface for the full interface definition.

Policy configuration

The third argument to stream_create is an optional JSON policy object that overrides execution defaults. Pass NULL to use all defaults.

{
  "max_retries": 5,
  "base_backoff_ms": 100,
  "window_grow_factor": 1.5,
  "window_shrink_factor": 0.5,
  "min_window_ms": 60000,
  "max_window_ms": 86400000,
  "prefetch_depth": 1
}

window_grow_factor and window_shrink_factor only affect the datadog_metrics adapter; they are ignored by generic_rest, openai, and anthropic.