Tool-Using ReAct

react is for tasks where the model should choose and chain tools. It still lives inside the behavior tree as one leaf.

from jdsl import react, root, tool

@tool
def distance_km(origin: str, destination: str) -> int:
    """Road distance between two cities."""
    ...

@tool
def drive_hours(km: int) -> float:
    """Driving time in hours."""
    ...

@tool
def fuel_cost(km: int) -> float:
    """Fuel cost for the trip."""
    ...

skill = (
    root("Trip", system="Use tools for every number.")
    .model("deepseek-chat")
    .do(react("request -> answer", tools=[distance_km, drive_hours, fuel_cost], max_steps=8))
)

Run the example:

uv run jdsl run examples/trip.py \
  -i request="Driving Nairobi to Mombasa - how long and how much fuel?"

Tool Schemas

@tool keeps the callable usable from Python and attaches metadata for the model. react derives a JSON schema from the function signature:

Python hint	Tool schema
str	string
int	integer
float	number
bool	boolean
list[T], tuple, set	array

Arguments without defaults are required. Unannotated arguments fall back to strings.

Failure and Recovery

If the model calls an unknown tool, jdsl sends an error observation back into the loop. If a tool raises, jdsl also sends an error observation rather than crashing the whole react leaf. The leaf fails only when no final answer arrives within max_steps or the final answer is empty.

Use a selector outside react when you want a deterministic fallback:

sel(
    react("request -> answer", tools=[search, fetch], max_steps=6),
    act(fallback_answer),
)