Skip to content
MicroResolve

Python Quickstart

MicroResolve ships a Python package (microresolve) with pre-built wheels. No build step, no native compiler needed.

Install

Terminal window
pip install microresolve

Requires Python 3.8+. Pre-built wheels for Linux, macOS, and Windows (x86_64 and arm64).

Basic usage

from microresolve import MicroResolve


engine = MicroResolve()
ns = engine.namespace("support")


ns.add_intent("cancel_order", [
    "cancel my order",
    "I want to cancel",
    "stop my order",
])
ns.add_intent("track_order", [
    "where is my package",
    "track my order",
    "shipping status",
])


matches = ns.resolve("I need to cancel")
for m in matches:
    print(f"{m['id']}: {m['score']:.2f}")
# cancel_order: 0.91

Persist to disk

from microresolve import MicroResolve


engine = MicroResolve(data_dir="~/.local/share/microresolve")
ns = engine.namespace("support")
# ... add intents and phrases ...
engine.flush()

On the next startup, namespaces are auto-loaded from data_dir.

Custom threshold per namespace

ns_security = engine.namespace_with("security", threshold=1.3)

Override threshold per call

matches = ns.resolve("cancel my order and get a refund", threshold=0.3, gap=1.5)

Add phrases at runtime

result = ns.add_phrase("cancel_order", "abort my order")
print(result["added"])      # True
print(result["redundant"])  # False if new


# Specify language explicitly
ns.add_phrase("cancel_order", "取消我的订单", lang="zh")

Correct a wrong classification

# "stop shipment" classified as cancel_order, should be track_order
ns.correct("stop shipment", wrong="cancel_order", correct="track_order")


# Teach from a miss (engine returned nothing)
ns.correct("where did my parcel go", wrong=None, correct="track_order")

List namespaces and intents

ns_ids = engine.namespaces()
intent_ids = ns.intent_ids()
print(ns.intent_count())

LangChain / LLM fallback pattern

from microresolve import MicroResolve


engine = MicroResolve(data_dir="~/.local/share/microresolve")
ns = engine.namespace("support")


def handle(query: str) -> str:
    matches = ns.resolve(query)
    if matches and matches[0]["score"] > 0.7:
        return handle_intent(matches[0]["id"], query)
    return llm_fallback(query)

Next