Create a BTREE scalar index on the ``scope`` column if not present. A scalar index lets LanceDB skip a full table scan when filtering by

Run the core execution logic of the task.

Execute the first tool from the tool_uses list and return its result. This is used when available_functions is provided, to directly execute

Execute a project script through uv. Args: script_name: The project script to run. entity_type: The user-facing entity being run.

Extract token usage and response metadata from Azure response.

Extract and parse all built-in tool outputs from Responses API. Parses web_search, file_search, code_interpreter, computer_use, and r

Extract raw finish_reason (``stopReason``) from a Bedrock Converse response dict. Defensive — returns (None, None) on any failure. Be

Extract reasoning items with encrypted content from response. Used for ZDR (Zero Data Retention) compliance to capture encrypted reas

Extract finish_reason and response_id from an OpenAI Responses API ``Response`` object. The Responses API exposes ``status`` rather th

Format feedback as a message.

Get or create the async OpenAI client.

Return the path to the version cache file, creating the dir if needed.

Get or create the S3 client.

Get or create the Anthropic client.

Get client parameters for compatibility with base class. Note: This method is kept for compatibility but the Google Gen AI SDK uses a

Extract collection creation parameters from kwargs.

Generate a search query for the knowledge base based on the task description.

Retrieve a memory prompt from the i18n translations. Args: key: The prompt key under the "memory" section. Returns: The prom

Get unique provider names from cache entries. Args: cache: The upload cache. Returns: Set of provider names.

Extract user context for tracing.

Get video duration in seconds using av if available. Args: content: Raw video bytes. content_type: Optional MIME type for format

Handle a tool call from the LLM. Args: tool_calls: List of tool calls from the LLM available_functions: Dict of avail

Import docling submodules lazily and cache the result. Raises: ImportError: If the docling package is not installed.

Increment the usage count.

Initialize trace batch - auto-enable ephemeral for first-time users. Args: user_context: User context information. ex

Inject merged metadata into call details.

Interpolates the inputs in the tasks and agents.

Type guard to check if the embedding function is async. Args: func: The embedding function to check. Returns: True if the fu

Check if a version string represents a pre-release.

Check if an object is a Pydantic model class. This distinguishes between Pydantic model classes that support structured outputs (have model_j

Return True for regular files, excluding symlinks to files.

Create a TraceBatchManager with a pre-set trace_batch_id.

Build a MethodExecutionStartedEvent with explicit ids/sequence.

Build a 404 message that says what to do about it. OpenAI's own wording ("This is not a chat model") doesn't make the fix obvious, so

Parse and validate the input arguments against the schema. Args: raw_args: The raw arguments to parse, either as a string or dict

Reconstruct a MemoryRecord from a Qdrant payload.

Poll the token endpoint until authentication completes or times out.

Capture execution context and sync checkpoint fields on each entity. Args: root: List of entities to prepare for serialization.

Preprocess structured data to handle common LLM output format issues. Some models (especially Claude on Bedrock) may return array fields as m

Resolve user input to a full repo name, or None if not found.

Retry an operation with exponential backoff. Args: operation: Async operation to retry. timeout: Operation timeout in

Write *content* to *filename*, confined to the tool's sandbox.

Run the crew with the trigger payload using the run_with_trigger method.

Set task identity and fingerprint data on an event.

Setup and invoke before_llm_call hooks for the executor context. Args: executor_context: The executor context to setup the hooks for.

Set up messages for the agent execution.

Whether to run PlannerObserver LLM calls after each step. Explicit ``observe_steps=False`` disables observation at any effort level.

Whether ``point`` must not be dispatched for this flow. Per point rather than per flow: an internal flow that is itself the whole run

The single exported span with ``name``, failing loudly if it is not unique.

Check if the model supports native structured outputs. Covers Claude Fable 5, Opus 5, Sonnet 5, Opus 4.8 and the 4.5-era models (Sonnet 4.5,

Build an in-memory tar archive via `build(tf)` and return it for reading.

Parse streamed observation JSON and update the exact step it names.

Remove key tracking for a provider.

Convert a UUID string to a stable Qdrant point ID. Falls back to hashing for non-UUID strings.

Validate keyword arguments against args_schema if present. Args: kwargs: The keyword arguments to validate. Returns:

Write a project-local ./skills/<name>/SKILL.md under *tmp_path*.

Call the LLM with the given messages. Args: messages: Input messages for the LLM. Can be a string or list of

Get merged crew and task files asynchronously. Task files override crew files with the same name. Args: crew_id: Unique identifier f

Get all cached uploads for a provider. Args: provider: The provider name. Returns: List of cached uploads fo

Query across all knowledge sources asynchronously. Args: query: List of query strings. results_limit: Maximum number

Execute a tool asynchronously. Args: calling: The tool calling information. tool_string: The raw tool string from the

Rebind the runtime ``Memory`` dependency after restore. Required after deserializing from a checkpoint, since the live ``Memory`` can

Override streaming for ``llm`` within the current call scope.

Restore the execution uuid that was active before ``token`` was issued. ``token`` is required so this cannot wipe an outer kickoff's uuid.

Remove the stored token file (no-op if absent).

Clear all user settings.

Nearest candidate name to a likely typo, if one is close enough.

Start the MCP server process and establish connection. Returns: Self for method chaining. Raises: Connection

Printable text content for chunk-like consumers.