Adaline

The main client class for interacting with the Adaline platform. Provides methods to fetch deployments, manage cached deployment controllers with background refresh, and initialize monitoring.

Constructor

from adaline.main import Adaline

adaline = Adaline(
    api_key="your-api-key",        # optional, defaults to ADALINE_API_KEY env var
    host="https://api.adaline.ai/v2",  # optional, defaults to ADALINE_BASE_URL env var
    debug=False                     # optional, enables DEBUG logging
)

All constructor parameters are keyword-only.

Parameters

str | None

Your Adaline API key. Defaults to the ADALINE_API_KEY environment variable.

str | None

The base URL for the Adaline API. Falls back to the ADALINE_BASE_URL environment variable, then https://api.adaline.ai/v2.

bool

default:"False"

If True, enables DEBUG-level logging on the adaline logger with a StreamHandler that outputs [Adaline] LEVEL: message.

Methods

get_deployment

Fetches a specific prompt deployment by prompt ID and deployment ID. This is an async method.

deployment = await adaline.get_deployment(
    prompt_id="your-prompt-id",
    deployment_id="your-deployment-id"
)

Parameters

str

required

The unique ID of the prompt.

str

required

The specific deployment ID.

Returns: A Deployment object. Raises: ApiException if the API call fails (4xx errors fail immediately, 5xx errors are retried).

get_latest_deployment

Fetches the latest deployment for a prompt in a specific environment. This is an async method.

deployment = await adaline.get_latest_deployment(
    prompt_id="your-prompt-id",
    deployment_environment_id="your-environment-id"
)

Parameters

str

required

The unique ID of the prompt.

str

required

The deployment environment ID.

Returns: The latest Deployment object for the given environment. Raises: ApiException if the API call fails.

init_latest_deployment

Initializes a cached latest deployment with automatic background refresh. Fetches the latest deployment immediately, then starts a background loop that refreshes the cache at the given interval. This is an async method.

controller = await adaline.init_latest_deployment(
    prompt_id="your-prompt-id",
    deployment_environment_id="your-environment-id",
    refresh_interval=60,
    max_continuous_failures=3
)

# Retrieve cached deployment (no API call unless forced)
deployment = await controller.get()

# Force a refresh from the API
deployment = await controller.get(force_refresh=True)

# Check background refresh status
status = controller.get_background_status()
# {"stopped": False, "consecutive_failures": 0, "last_error": None, "last_refreshed": datetime}

# Stop the background refresh
await controller.stop()

Parameters

str

required

The unique ID of the prompt.

str

required

The deployment environment ID.

int

default:"60"

Seconds between background refreshes. Clamped to [1, 600].

int

default:"3"

Number of consecutive failures before the background loop stops itself.

Returns: A Controller instance for retrieving and managing the cached deployment. Raises: ApiException if the initial fetch fails.

Controller

The Controller class returned by init_latest_deployment provides:

Method	Signature	Description
`get`	`async get(force_refresh: bool = False) -> Optional[Deployment]`	Returns the cached deployment. If `force_refresh=True`, bypasses the cache.
`stop`	`async stop()`	Cancels the background refresh task and clears the cache entry.
`get_background_status`	`get_background_status() -> dict`	Returns a snapshot: `{"stopped", "consecutive_failures", "last_error", "last_refreshed"}`.

init_evaluation_results

Initializes a cached, polling fetcher for evaluation results. Mirrors init_latest_deployment but targets EvaluationsClient.get_results. Useful when an evaluation is still running server-side and you want to surface partial results without writing your own polling loop — the same query (pagination + filters) is replayed on every refresh. This is an async method.

results = await adaline.init_evaluation_results(
    prompt_id="your-prompt-id",
    evaluation_id="your-evaluation-id",
    grade="fail",
    expand="row",
    sort="score:desc",
    limit=50,
    refresh_interval=30,
    max_continuous_failures=3,
)

# Retrieve cached page (no API call unless forced)
page = await results.get()

# Force a refresh
page = await results.get(force_refresh=True)

# Check background refresh status
status = results.get_background_status()

# Stop the background refresh when you're done
await results.stop()

Parameters

str

required

The unique ID of the prompt.

str

required

The unique ID of the evaluation.

str | None

Filter by grade: "pass", "fail", or "unknown".

str | None

Pass "row" to include the underlying dataset row in each result.

str | None

Sort order: "createdAt:asc", "createdAt:desc", "score:asc", or "score:desc".

int | None

Page size.

str | None

Pagination cursor from a previous response.

int

default:"60"

Seconds between background refreshes. Clamped to [1, 600].

int

default:"3"

Consecutive failures before the background loop stops itself.

Returns: An EvaluationResultsController exposing get(force_refresh=False), get_background_status(), and stop() — identical in shape to the deployment controller. Raises: ApiException if the initial fetch fails.

init_monitor

Initializes and returns a new Monitor instance for logging traces and spans. This is a synchronous method.

monitor = adaline.init_monitor(
    project_id="your-project-id",
    flush_interval_seconds=1,
    max_buffer_size=1000,
    default_content=None
)

Parameters

str

required

Unique project identifier for grouping traces and spans.

int

default:"1"

Interval (in seconds) at which the buffer is automatically flushed.

int

default:"1000"

Maximum number of buffered entries before oldest items are dropped.

LogSpanContent | None

Default content attached to spans when none is provided. See LogSpanContent for available types. Defaults to an LogSpanOtherContent with empty input/output.

Namespace clients

Seven namespace clients are attached to every Adaline instance. Each wraps the corresponding autogen API with retry-on-5xx, abort-on-4xx, and keyword-only arguments:

Attribute	Client	Covers
`adaline.datasets`	DatasetsClient	Datasets (+ `.rows`, `.columns` sub-clients)
`adaline.prompts`	PromptsClient	Prompts (+ `.draft`, `.playgrounds`, `.evaluators`, `.evaluations` sub-clients)
`adaline.providers`	ProvidersClient	configured LLM providers
`adaline.models`	ModelsClient	models available across providers
`adaline.projects`	ProjectsClient	list / get / update workspace projects
`adaline.logs`	LogsClient	read-side log access (+ `.traces`, `.spans` sub-clients)

Raw escape hatches are also exposed:

adaline.deployments_api — raw DeploymentsApi from adaline_api
adaline.logs_api — raw LogsApi (used internally by Monitor)

If you need identical retry behavior on an arbitrary call, use the exported with_retry helper:

from adaline.main import Adaline
from adaline.clients import with_retry

adaline = Adaline()

response = await with_retry(lambda: adaline.projects.list())

Returns: A Monitor instance.

Get started

API

SDKs

Proxy

Adaline

Adaline

Constructor

Parameters

Methods

get_deployment

Parameters

get_latest_deployment

Parameters

init_latest_deployment

Parameters

Controller

init_evaluation_results

Parameters

init_monitor

Parameters

Namespace clients

​Adaline

​Constructor

​Parameters

​Methods

​get_deployment

​Parameters

​get_latest_deployment

​Parameters

​init_latest_deployment

​Parameters

​Controller

​init_evaluation_results

​Parameters

​init_monitor

​Parameters

​Namespace clients

Adaline

Constructor

Parameters

Methods

get_deployment

Parameters

get_latest_deployment

Parameters

init_latest_deployment

Parameters

Controller

init_evaluation_results

Parameters

init_monitor

Parameters

Namespace clients