ovos-ddg-plugin

DuckDuckGo Instant Answers plugin for OpenVoiceOS.

Provides factual Q&A, structured infobox lookups, and image URLs via the DDG Instant Answers API. No API key required.

Implements the opm.agents.retrieval and opm.agents.toolbox entry points — usable both as a standalone Python library and as an OVOS agent plugin.

---

Installation

pip install ovos-ddg-plugin

Legacy users: the old ovos-ddg-solver-plugin package remains on PyPI. New code should import from ovos_ddg_plugin.

---

Standalone usage

Text answers

from ovos_ddg_plugin import DuckDuckGoRetrievalEngine

ddg = DuckDuckGoRetrievalEngine()

# Returns up to k (sentence, score) tuples, scored 0.9 → 0.8 → 0.7 …
results = ddg.query("who is Isaac Newton", lang="en-us", k=3)
for sentence, score in results:
    print(f"[{score:.1f}] {sentence}")
# [0.9] Sir Isaac Newton was an English polymath active as a mathematician …
# [0.8] He was a key figure in the Scientific Revolution …
# [0.7] His book Philosophiæ Naturalis Principia Mathematica …

# Infobox field queries return a single result at score 0.9
results = ddg.query("when was Stephen Hawking born", lang="en-us")
print(results)
# [('January 8, 1942', 0.9)]

Infobox

infobox, related = ddg.get_infobox("Stephen Hawking", lang="en-us")
from pprint import pprint
pprint(infobox)
# {'born': 'January 8, 1942',
#  'died': 'March 14, 2018',
#  'known_for': 'Hawking radiation, A Brief History of Time …',
#  'alma_mater': 'University of Oxford; University of Cambridge',
#  ...}
print(related[:3])
# ['A Brief History of Time', 'Cambridge University', 'General relativity']

Image URL

url = ddg.get_image("Eiffel Tower", lang="en-us")
print(url)
# https://duckduckgo.com/i/...jpg  (or None if DDG has no image)

---

Agent tools

DuckDuckGoToolbox exposes three tools for OVOS agent pipelines:

Tool	Description
search_duckduckgo	Best text answer (infobox field or abstract sentence)
duckduckgo_infobox	Full structured infobox + related topics for an entity
duckduckgo_image	Image URL for a topic, or null

from ovos_ddg_plugin import DuckDuckGoToolbox, SearchDuckDuckGoArgs, DDGInfoboxArgs, DDGImageArgs

tb = DuckDuckGoToolbox()

answer = tb.search_ddg(SearchDuckDuckGoArgs(query="Marie Curie birthdate", lang="en-us"))
print(answer.result)
# November 7, 1867

box = tb.ddg_infobox(DDGInfoboxArgs(query="Marie Curie", lang="en-us"))
print(box.infobox.get("born"))
# November 7, 1867
print(box.related_topics[:2])
# ['Radioactivity', 'Nobel Prize in Physics']

img = tb.ddg_image(DDGImageArgs(query="Marie Curie", lang="en-us"))
print(img.url)
# https://duckduckgo.com/i/...jpg

---

Infobox intent matching

When query() receives a natural-language question, it runs it through a bank of Padacioso intent files to check whether the question targets a specific infobox field. If a match is found the engine fetches the infobox for the extracted entity and returns the exact field value at score 0.9, bypassing the abstract entirely.

How it works

1. Intent detection — _match_infobox_intents() calls Padacioso's calc_intents() (plural) and returns all candidate matches sorted by confidence, each as (intent_name, keyword, entity_type).

2. Entity type — the slot name used in the intent file becomes the entity_type ("person", "movie", "place", "animal", etc.). This contextualises the keyword for downstream use.

3. Infobox-based disambiguation — when two intents tie (e.g. length and running_time both match "how long is Amazon"), query() walks the candidate list, fetches the DDG infobox for the extracted keyword, and returns the first candidate whose field is non-empty in the infobox. The entity type associated with the winning slot does not need to be known ahead of time — the infobox result decides.

4. Field aliases — DDG does not always name its keys the same as the intent. FIELD_ALIASES maps intent names to the list of DDG keys to try in order (e.g. alma_mater → ["education"], resting_place → ["resting_place", "burial", "burial_place"]).

5. Fallback — if no candidate's field is present in the infobox, query() falls through to the full-text abstract.

Supported intents

79 intents across 10 locales. Entity-typed slots ({person}, {movie}, {place}, …) are used instead of a generic {keyword} wherever the context constrains the entity type.

Person / biography

Intent	Sample pattern
born	when was {person} born
died	when did {person} die
age	how old is {person}
age_at_death	how old was {person} when they died
known_for	what is {person} known for
father / mother / children	who is {person} father
spouse / partner	who is {person} married to
nickname	what is {person} nickname
citizenship	what is {person} nationality
net_worth	what is {person} net worth
height	how tall is {person}
resting_place	where is {person} buried
notable_work	what is {person} most famous work
notable_awards	what awards has {person} won

Person / career & academic

Intent	Sample pattern
occupation	what does {person} do
fields	what field does {person} work in
institutions	what institution does {person} work for
alma_mater	where did {person} go to university
education	where did {person} study
thesis	{person} doctoral thesis
doctoral_students	who were {person} doctoral students
movement	what movement is {person} associated with
religion	what religion is {person}
official_website	{person} official website
years_active	when was {person} active
political_party	what party does {person} belong to
predecessor / successor	who came before / after {person}
coached_by	who coaches {person}
sport	what sport does {person} play

Film / media

Intent	Sample pattern
director	who directed {movie}
starring	who stars in {movie}
released	when was {movie} released
running_time	how long is {movie}
budget	what was the budget of {movie}
box_office	how much did {movie} earn
produced_by	who produced {movie}
written_by	who wrote the screenplay for {movie}
composers	who composed the music for {movie}
original_title	what is the original title of {movie}
genre	what genre is {movie}
language	what language is {book} in

Book / literature

Intent	Sample pattern
author	who wrote {book}
publisher	who published {book}

Music

Intent	Sample pattern
label	what record label is {artist} on
members	who are the members of {group}

Game / software

Intent	Sample pattern
developer	who developed {software}
platforms	what platforms is {software} on
license	what license is {software} under

Organisation / company

Intent	Sample pattern
founded	when was {organization} founded
founders	who founded {organization}
industry	what industry is {company} in
ceo	who is the CEO of {company}
owner	who owns {keyword}
revenue	what is {company} revenue
number_of_employees	how many employees does {company} have
headquarters	where is {company} headquartered
head_coach	who is the head coach of {team}
stadium	what stadium does {team} play in
championships	how many championships has {team} won

Geography / place

Intent	Sample pattern
capital	what is the capital of {place}
country	what country is {place} in
location	where is {place} located
area	what is the area of {place}
population	how many people live in {place}
elevation	how high is {mountain}
length	how long is {place}

Buildings / structures

Intent	Sample pattern
opened	when was {place} opened
designed_by	who designed {place}

Biology / nature

Intent	Sample pattern
scientific_name	what is the scientific name of {animal}
habitat	where does {animal} live
diet	what does {animal} eat
conservation_status	is {animal} endangered

Generic

Intent	Sample pattern
other_names	what are {keyword} other names
inventor	who invented {keyword}

Supported locales

Intent files ship for 10 locales, each with 6–12 sample patterns per intent:

Locale	Language
en-US	English
ca-ES	Catalan
da-DK	Danish
de-DE	German
es-ES	Spanish
eu-ES	Basque
fr-FR	French
gl-ES	Galician
it-IT	Italian
pt-PT	Portuguese

Matching is case-insensitive and possessive 's is stripped before matching so "Darwin's father" and "Darwin father" both work.

Adding or extending intent files

Intent files live in ovos_ddg_plugin/locale/<locale>/. Each file is named <intent>.intent and contains one Padacioso pattern per line:

who directed {movie}
{movie} director
who made {movie}

Slot names constrain the entity type and are used as the entity_type value returned by _match_infobox_intents(). Use specific types ({person}, {movie}, {place}, {animal}, {company}, {software}, {book}, {artist}, {group}, {organization}, {mountain}, {team}) instead of the generic {keyword} wherever the context restricts what the entity can be. Use {keyword} only for truly open-ended slots (e.g. inventor, owner, other_names).

If the DDG infobox key differs from the intent name, add an entry to FIELD_ALIASES in __init__.py:

FIELD_ALIASES["my_intent"] = ["ddg_key_1", "ddg_key_2"]

---

Supported locales (DDG API)

The engine maps any BCP-47 tag to the closest DDG locale code via langcodes. Infobox date fields (born, died) are formatted in the requested language using ovos-date-parser.

---

Configuration

All keys are optional and read from the OVOS plugin config block for ovos-ddg-plugin:

Key	Default	Description
keyword_extractor	"ovos-rake-keyword-extractor"	OPM keyword extractor plugin used when a direct query returns no result

---

License

Apache 2.0 — see LICENSE.