Build PostgreSQL connection URL from connection information. Args: connection_info: Connection information from Render A

Calculate final metrics from running totals. Args: metrics: Dictionary of running metrics num_samples: Total

Evaluate search performance on a dataset. Args: dataset_artifact: Name of the dataset artifact to evaluate d

Async wrapper function that evaluates a website and updates the database with results. This function handles the complete evaluation lifecyc

Execute SQL query synchronously using SQLAlchemy. Args: connection_url: PostgreSQL connection URL sql: SQL q

Fetch app and function data from the database. Returns: DataFrame containing app and function information

Find the rank (1-based) of the expected function in the search results. Args: results: List of search results

Generate synthetic intents. Args: dataset_artifact: Name of the artifact to save the dataset to dataset_file

Generate fuzzy input arguments for a function with LLM.

Generate a single intent using OpenAI's API. Args: prompt: The prompt to send to the model Returns:

Get API key credentials from linked account or use app's default credentials if no linked account's API key is found and if the app has a def

Get the app connector class. Returns: The app connector class. Raises: NoImplementationFound: If th

Get the actual client IP if the server is running behind a proxy.

Get the error message from the response or fallback to the error message from the HTTPStatusError. Usually the response json contains more det

a somewhat no-op function, but we keep it for consistency.

Get OAuth2 credentials from linked account or app's default credentials. If the access token is expired, it will be refreshed.

Get PostgreSQL connection information from Render API. Args: postgres_id: The ID of the PostgreSQL instance Ret

Get the response data from the response. If the response is json, return the json data, otherwise fallback to the text.

Load a dataset from a W&B artifact. Args: artifact_name: Name of the W&B artifact dataset_filename: Filename

Log dataset statistics to wandb. Args: df: DataFrame containing the generated dataset

Determines if the function embedding should be regenerated based on changes in the fields used for embedding (name, description, parameters).

Render a Jinja2 template with the provided secrets and return as string.

Save the dataset as a wandb artifact. Args: df: DataFrame containing the generated dataset dataset_artifact:

Send a search request to the API and measure response time. Args: intent: The search query/intent limit: Max

Update running metrics with results from a single evaluation. Args: metrics: Dictionary of running metrics to update

Validate URL format and security to prevent SSRF attacks. TODO: the best way to prevent SSRF attacks is actually put this agent execution in

Convert a list of ClientMessage objects to a list of OpenAI messages. Args: messages: A list of ClientMessage objects Returns:

Create a new plan.

Create authorization URL for user to authorize your application Args: redirect_uri: The redirect URI of the OAuth2 clien

Batch creates functions in the database. Generates embeddings for each new function and calls the CRUD layer for creation. Returns a list

Deletes a credential for a specific domain. Function name: AGENT_SECRETS_MANAGER__DELETE_CREDENTIAL_FOR_DOMAIN Args:

Evaluate search performance on a dataset of synthetic intents. Args: dataset: DataFrame containing synthetic intents and

Execute the function based on end-user input and security credentials. Input validation, default values injection, and security crede

Execute a function with the given parameters. Args: db_session: Database session project: Project object agent: Agen

Exchange authorization code for access token Args: redirect_uri: The redirect URI of the OAuth2 client code:

Test function execution with GPT-generated inputs.

Generate synthetic intents and save them. Args: dataset_artifact: Name of the artifact to save the dataset to

Generate a random code verifier for OAuth2