ReasoningChain

ReasoningChain is the main public API: a list of steps plus execution settings. It runs them through the DAG executor and serialises to JSON for reuse.

Constructor

ReasoningChain(steps, max_workers=3, ...)

Parameter	Type	Default	Purpose
steps	Sequence[StepDescription...]	— (required)	The steps to run.
max_workers	int | str	3	Parallel worker pool size. "auto" sizes it for you.
enable_progress	bool	False	Emit progress logging.
search_config	ContextSearchConfig | None	None	Context-extraction strategy (substring / vector).
metrics	list[MetricBase]	[]	Chain-level metrics.
timeout	float | None	None	Chain-level timeout in seconds.
replan_policy	ReplanPolicy | None	None	RE-PLAN policy.
default_llm_config	LLMStepConfig | None	None	Default LLM config for all LLM steps.
memory_schema	dict | None	None	Write-time memory validation schema.
metadata	dict | None	None	Arbitrary metadata stored on the chain.

(Also: prompt_template, trace_name, session_id, step_groups, max_injections.)

Running a chain

result = chain.execute(context)             # synchronous
result = await chain.execute_async(context) # async

For step-by-step streaming, use stream_async. See async execution for parallelism, callbacks, and timeouts.

Serialization

Chains round-trip to JSON so you can save and reload them:

chain.save("my_chain.json")              # write to disk
loaded = ReasoningChain.load("my_chain.json")

d = chain.to_dict();  ReasoningChain.from_dict(d)
s = chain.to_json();  ReasoningChain.from_json(s)

from_dict runs full validation (cycles, dependency references, reference-syntax warnings).

Other chain methods

• chain.estimate_cost(pricing=...) — dry-run token/USD projection before running.
• chain.to_mermaid() — render the DAG as a Mermaid diagram.
• chain.reflect(...) — analyse a completed run.

Ways to build a chain

• Directly — pass a list of typed step descriptions (this page).
• ChainBuilder — a fluent builder.
• ChainBuilder.from_description — generate a chain from natural language.