1. Removes LLM 'thinking' traces (e.g., <think> tags). 2. Extracts the first JSON object '{ ... }' from the text. 3. Parses it into a dic

The "Sensible Default" prompt builder. Automatically constructs a highly structured, strict identity prompt. Users can utiliz

Executes the parallel reasoning workflow. 1. Launches all agents simultaneously in separate threads. 2. Collects raw

Helper method to standardize how agent outputs are presented to the LLM.

Helper function to call OpenAI directly for deterministic analysis in Phase 1.

Standardizes the agent's output into a clean string for the Aggregator. Automatically serializes Dictionaries, Dataclasses, and Pydan

The core reasoning logic. Use `self._build_prompt(problem_data)` to get your strict isolated prompt, or completely ignore it

Initializes the Agent. Args: role: The specific persona (e.g., "Senior Financial Analyst"). goal: Wh

Initializes the Aggregator. Args: role: The persona of the aggregator. goal: The overarching objecti

Initialize the engine with workers and a boss. Args: agents (List[Agent]): A list of specialist agents (workers)

Helper function to call the local Ollama model via LiteLLM.

Helper function to call OpenAI via LiteLLM for the Aggregator.

Takes the results from all parallel agents. Returns the final verdict (String, Dict, or Object). Use `self._format_reports(ag

A standard, model-agnostic callable to pass into the agents and aggregators. Octochains doesn't force you into a specific LLM wrapper. You de

Tests that the _build_prompt helper generates the strict identity prompt.

Tests that format_output correctly invokes Pydantic's model_dump_json.

Tests that a failing agent does not crash the entire engine workflow.

Tests standard text-based execution.

Tests that the Engine cleanly handles Python dictionary objects without crashing.

Tests that extra keys are ignored gracefully if model_config allows it.

Tests that Pydantic ValidationError is properly wrapped in a ValueError.

Tests that the parser correctly extracts and maps valid JSON to a Pydantic model.

Tests that reasoning model artifacts (<think>) are cleanly stripped.

Standardized helper for JSON serialization.

Standardized helper for JSON serialization.

Standardized helper for JSON serialization.