Improve Online Search: Parallelize Search, Use Jina Reader API by default (#832)

- Overview
  Khoj wil be able to do online search out of the box, even for self-hosted users
  - Default to Jina search, reader API when no Serper.dev, Olostep API keys
  - Run online searches in parallel to process multiple queries faster

- Details
  - Jina provides a [reader API](https://github.com/jina-ai/reader) for online search and web page reading
     It requires no API key. This provides a good default to enable 
     online search for self-hosted readers requiring no additional setup. 

  - Jina search API also returns webpage contents with the results, so
     just use those directly when Jina Search API used instead of
     trying to read webpages separately. The extract relevant content from
     webpage step using a chat model is still used from the
    `read_webpage_and_extract_content' func in this case.

  - Parse search results from Jina search API into same format as
     Serper.dev for accurate rendering of online references by clients

  - Run online searches in parallel with AsyncIO to process multiple queries faster
This commit is contained in:
Debanjum
2024-07-02 17:44:51 +05:30
committed by GitHub
6 changed files with 129 additions and 97 deletions

View File

@@ -1,17 +1,21 @@
# Online Search
By default, Khoj will try to infer which information-sourcing tools are required to answer your question. Sometimes, you'll have a need for outside questions that the LLM's knowledge doesn't cover. In that case, it will use the `online` search feature.
Khoj will research on the internet to ground its responses, when it determines that it would need fresh information outside its existing knowledge to answer the query. It will always show any online references it used to respond to your requests.
For example, these queries would trigger an online search:
By default, Khoj will try to infer which information sources, it needs to read to answer your question. This can include reading your documents or researching information online. You can also explicitly trigger an online search by adding the `/online` prefix to your chat query.
Example queries that should trigger an online search:
- What's the latest news about the Israel-Palestine war?
- Where can I find the best pizza in New York City?
- Deadline for filing taxes 2024.
- /online Deadline for filing taxes 2024.
- Give me a summary of this article: https://en.wikipedia.org/wiki/Haitian_Revolution
Try it out yourself! https://app.khoj.dev
## Self-Hosting
The general online search function currently requires an API key from Serper.dev. You can grab one here: https://serper.dev/, and then add it as an environment variable with the name `SERPER_DEV_API_KEY`.
Online search works out of the box even when self-hosting. Khoj uses [JinaAI's reader API](https://jina.ai/reader/) to search online and read webpages by default. No API key setup is necessary.
Without any API keys, Khoj will use the `requests` library to directly read any webpages you give it a link to. This means that you can use Khoj to read any webpage that you have access in your local network.
To improve online search, set the `SERPER_DEV_API_KEY` environment variable to your [Serper.dev](https://serper.dev/) API key. These search results include additional context like answer box, knowledge graph etc.
For advanced webpage reading, set the `OLOSTEP_API_KEY` environment variable to your [Olostep](https://www.olostep.com/) API key. This has a higher success rate at reading webpages than the default webpage reader.

View File

@@ -1,5 +1,4 @@
import logging
import os
from threading import Thread
from typing import Dict
@@ -40,7 +39,7 @@ def completion_with_backoff(
client: openai.OpenAI = openai_clients.get(client_key)
if not client:
client = openai.OpenAI(
api_key=openai_api_key or os.getenv("OPENAI_API_KEY"),
api_key=openai_api_key,
base_url=api_base_url,
)
openai_clients[client_key] = client
@@ -102,7 +101,7 @@ def llm_thread(g, messages, model_name, temperature, openai_api_key=None, api_ba
client_key = f"{openai_api_key}--{api_base_url}"
if client_key not in openai_clients:
client: openai.OpenAI = openai.OpenAI(
api_key=openai_api_key or os.getenv("OPENAI_API_KEY"),
api_key=openai_api_key,
base_url=api_base_url,
)
openai_clients[client_key] = client

View File

@@ -2,11 +2,11 @@ import asyncio
import json
import logging
import os
import urllib.parse
from collections import defaultdict
from typing import Callable, Dict, List, Optional, Tuple, Union
import aiohttp
import requests
from bs4 import BeautifulSoup
from markdownify import markdownify
@@ -23,6 +23,10 @@ logger = logging.getLogger(__name__)
SERPER_DEV_API_KEY = os.getenv("SERPER_DEV_API_KEY")
SERPER_DEV_URL = "https://google.serper.dev/search"
JINA_READER_API_URL = "https://r.jina.ai/"
JINA_SEARCH_API_URL = "https://s.jina.ai/"
JINA_API_KEY = os.getenv("JINA_API_KEY")
OLOSTEP_API_KEY = os.getenv("OLOSTEP_API_KEY")
OLOSTEP_API_URL = "https://agent.olostep.com/olostep-p2p-incomingAPI"
OLOSTEP_QUERY_PARAMS = {
@@ -50,9 +54,6 @@ async def search_online(
custom_filters: List[str] = [],
):
query += " ".join(custom_filters)
if not online_search_enabled():
logger.warn("SERPER_DEV_API_KEY is not set")
return {}
if not is_internet_connected():
logger.warn("Cannot search online as not connected to internet")
return {}
@@ -61,27 +62,35 @@ async def search_online(
subqueries = await generate_online_subqueries(query, conversation_history, location)
response_dict = {}
for subquery in subqueries:
if subqueries:
logger.info(f"🌐 Searching the Internet for {list(subqueries)}")
if send_status_func:
await send_status_func(f"**🌐 Searching the Internet for**: {subquery}")
logger.info(f"🌐 Searching the Internet for '{subquery}'")
response_dict[subquery] = search_with_google(subquery)
subqueries_str = "\n- " + "\n- ".join(list(subqueries))
await send_status_func(f"**🌐 Searching the Internet for**: {subqueries_str}")
# Gather distinct web pages from organic search results of each subquery without an instant answer
webpage_links = {
organic["link"]: subquery
with timer(f"Internet searches for {list(subqueries)} took", logger):
search_func = search_with_google if SERPER_DEV_API_KEY else search_with_jina
search_tasks = [search_func(subquery) for subquery in subqueries]
search_results = await asyncio.gather(*search_tasks)
response_dict = {subquery: search_result for subquery, search_result in search_results}
# Gather distinct web page data from organic results of each subquery without an instant answer.
# Content of web pages is directly available when Jina is used for search.
webpages = {
(organic.get("link"), subquery, organic.get("content"))
for subquery in response_dict
for organic in response_dict[subquery].get("organic", [])[:MAX_WEBPAGES_TO_READ]
if "answerBox" not in response_dict[subquery]
}
# Read, extract relevant info from the retrieved web pages
if webpage_links:
if webpages:
webpage_links = [link for link, _, _ in webpages]
logger.info(f"🌐👀 Reading web pages at: {list(webpage_links)}")
if send_status_func:
webpage_links_str = "\n- " + "\n- ".join(list(webpage_links))
await send_status_func(f"**📖 Reading web pages**: {webpage_links_str}")
tasks = [read_webpage_and_extract_content(subquery, link) for link, subquery in webpage_links.items()]
tasks = [read_webpage_and_extract_content(subquery, link, content) for link, subquery, content in webpages]
results = await asyncio.gather(*tasks)
# Collect extracted info from the retrieved web pages
@@ -92,23 +101,24 @@ async def search_online(
return response_dict
def search_with_google(subquery: str):
payload = json.dumps({"q": subquery})
async def search_with_google(query: str) -> Tuple[str, Dict[str, List[Dict]]]:
payload = json.dumps({"q": query})
headers = {"X-API-KEY": SERPER_DEV_API_KEY, "Content-Type": "application/json"}
response = requests.request("POST", SERPER_DEV_URL, headers=headers, data=payload)
async with aiohttp.ClientSession() as session:
async with session.post(SERPER_DEV_URL, headers=headers, data=payload) as response:
if response.status != 200:
logger.error(await response.text())
return query, {}
json_response = await response.json()
extraction_fields = ["organic", "answerBox", "peopleAlsoAsk", "knowledgeGraph"]
extracted_search_result = {
field: json_response[field]
for field in extraction_fields
if not is_none_or_empty(json_response.get(field))
}
if response.status_code != 200:
logger.error(response.text)
return {}
json_response = response.json()
extraction_fields = ["organic", "answerBox", "peopleAlsoAsk", "knowledgeGraph"]
extracted_search_result = {
field: json_response[field] for field in extraction_fields if not is_none_or_empty(json_response.get(field))
}
return extracted_search_result
return query, extracted_search_result
async def read_webpages(
@@ -134,10 +144,13 @@ async def read_webpages(
return response
async def read_webpage_and_extract_content(subquery: str, url: str) -> Tuple[str, Union[None, str], str]:
async def read_webpage_and_extract_content(
subquery: str, url: str, content: str = None
) -> Tuple[str, Union[None, str], str]:
try:
with timer(f"Reading web page at '{url}' took", logger):
content = await read_webpage_with_olostep(url) if OLOSTEP_API_KEY else await read_webpage_at_url(url)
if is_none_or_empty(content):
with timer(f"Reading web page at '{url}' took", logger):
content = await read_webpage_with_olostep(url) if OLOSTEP_API_KEY else await read_webpage_with_jina(url)
with timer(f"Extracting relevant information from web page at '{url}' took", logger):
extracted_info = await extract_relevant_info(subquery, content)
return subquery, extracted_info, url
@@ -172,5 +185,41 @@ async def read_webpage_with_olostep(web_url: str) -> str:
return response_json["markdown_content"]
def online_search_enabled():
return SERPER_DEV_API_KEY is not None
async def read_webpage_with_jina(web_url: str) -> str:
jina_reader_api_url = f"{JINA_READER_API_URL}/{web_url}"
headers = {"Accept": "application/json", "X-Timeout": "30"}
if JINA_API_KEY:
headers["Authorization"] = f"Bearer {JINA_API_KEY}"
async with aiohttp.ClientSession() as session:
async with session.get(jina_reader_api_url, headers=headers) as response:
response.raise_for_status()
response_json = await response.json()
return response_json["data"]["content"]
async def search_with_jina(query: str) -> Tuple[str, Dict[str, List[Dict]]]:
encoded_query = urllib.parse.quote(query)
jina_search_api_url = f"{JINA_SEARCH_API_URL}/{encoded_query}"
headers = {"Accept": "application/json"}
if JINA_API_KEY:
headers["Authorization"] = f"Bearer {JINA_API_KEY}"
async with aiohttp.ClientSession() as session:
async with session.get(jina_search_api_url, headers=headers) as response:
if response.status != 200:
logger.error(await response.text())
return query, {}
response_json = await response.json()
parsed_response = [
{
"title": item["title"],
"content": item.get("content"),
# rename description -> snippet for consistency
"snippet": item["description"],
# rename url -> link for consistency
"link": item["url"],
}
for item in response_json["data"]
]
return query, {"organic": parsed_response}

View File

@@ -29,11 +29,7 @@ from khoj.processor.conversation.prompts import (
)
from khoj.processor.conversation.utils import save_to_conversation_log
from khoj.processor.speech.text_to_speech import generate_text_to_speech
from khoj.processor.tools.online_search import (
online_search_enabled,
read_webpages,
search_online,
)
from khoj.processor.tools.online_search import read_webpages, search_online
from khoj.routers.api import extract_references_and_questions
from khoj.routers.helpers import (
ApiUserRateLimiter,
@@ -767,22 +763,16 @@ async def websocket_endpoint(
conversation_commands.remove(ConversationCommand.Notes)
if ConversationCommand.Online in conversation_commands:
if not online_search_enabled():
conversation_commands.remove(ConversationCommand.Online)
# If online search is not enabled, try to read webpages directly
if ConversationCommand.Webpage not in conversation_commands:
conversation_commands.append(ConversationCommand.Webpage)
else:
try:
online_results = await search_online(
defiltered_query, meta_log, location, send_status_update, custom_filters
)
except ValueError as e:
logger.warning(f"Error searching online: {e}. Attempting to respond without online results")
await send_complete_llm_response(
f"Error searching online: {e}. Attempting to respond without online results"
)
continue
try:
online_results = await search_online(
defiltered_query, meta_log, location, send_status_update, custom_filters
)
except ValueError as e:
logger.warning(f"Error searching online: {e}. Attempting to respond without online results")
await send_complete_llm_response(
f"Error searching online: {e}. Attempting to respond without online results"
)
continue
if ConversationCommand.Webpage in conversation_commands:
try:
@@ -1067,18 +1057,10 @@ async def chat(
conversation_commands.remove(ConversationCommand.Notes)
if ConversationCommand.Online in conversation_commands:
if not online_search_enabled():
conversation_commands.remove(ConversationCommand.Online)
# If online search is not enabled, try to read webpages directly
if ConversationCommand.Webpage not in conversation_commands:
conversation_commands.append(ConversationCommand.Webpage)
else:
try:
online_results = await search_online(
defiltered_query, meta_log, location, custom_filters=_custom_filters
)
except ValueError as e:
logger.warning(f"Error searching online: {e}. Attempting to respond without online results")
try:
online_results = await search_online(defiltered_query, meta_log, location, custom_filters=_custom_filters)
except ValueError as e:
logger.warning(f"Error searching online: {e}. Attempting to respond without online results")
if ConversationCommand.Webpage in conversation_commands:
try:

View File

@@ -62,7 +62,6 @@ def test_offline_chat_with_no_chat_history_or_retrieved_content(client_offline_c
# ----------------------------------------------------------------------------------------------------
@pytest.mark.skipif(os.getenv("SERPER_DEV_API_KEY") is None, reason="requires SERPER_DEV_API_KEY")
@pytest.mark.chatquality
@pytest.mark.django_db(transaction=True)
def test_chat_with_online_content(client_offline_chat):
@@ -75,18 +74,18 @@ def test_chat_with_online_content(client_offline_chat):
response_message = response_message.split("### compiled references")[0]
# Assert
expected_responses = ["http://www.paulgraham.com/greatwork.html"]
expected_responses = [
"https://paulgraham.com/greatwork.html",
"https://www.paulgraham.com/greatwork.html",
"http://www.paulgraham.com/greatwork.html",
]
assert response.status_code == 200
assert any([expected_response in response_message for expected_response in expected_responses]), (
"Expected links or serper not setup in response but got: " + response_message
)
assert any(
[expected_response in response_message for expected_response in expected_responses]
), f"Expected links: {expected_responses}. Actual response: {response_message}"
# ----------------------------------------------------------------------------------------------------
@pytest.mark.skipif(
os.getenv("SERPER_DEV_API_KEY") is None or os.getenv("OLOSTEP_API_KEY") is None,
reason="requires SERPER_DEV_API_KEY and OLOSTEP_API_KEY",
)
@pytest.mark.chatquality
@pytest.mark.django_db(transaction=True)
def test_chat_with_online_webpage_content(client_offline_chat):
@@ -101,9 +100,9 @@ def test_chat_with_online_webpage_content(client_offline_chat):
# Assert
expected_responses = ["185", "1871", "horse"]
assert response.status_code == 200
assert any([expected_response in response_message for expected_response in expected_responses]), (
"Expected links or serper not setup in response but got: " + response_message
)
assert any(
[expected_response in response_message for expected_response in expected_responses]
), f"Expected response with {expected_responses}. But actual response had: {response_message}"
# ----------------------------------------------------------------------------------------------------

View File

@@ -61,7 +61,6 @@ def test_chat_with_no_chat_history_or_retrieved_content(chat_client):
# ----------------------------------------------------------------------------------------------------
@pytest.mark.skipif(os.getenv("SERPER_DEV_API_KEY") is None, reason="requires SERPER_DEV_API_KEY")
@pytest.mark.chatquality
@pytest.mark.django_db(transaction=True)
def test_chat_with_online_content(chat_client):
@@ -74,18 +73,18 @@ def test_chat_with_online_content(chat_client):
response_message = response_message.split("### compiled references")[0]
# Assert
expected_responses = ["http://www.paulgraham.com/greatwork.html"]
expected_responses = [
"https://paulgraham.com/greatwork.html",
"https://www.paulgraham.com/greatwork.html",
"http://www.paulgraham.com/greatwork.html",
]
assert response.status_code == 200
assert any([expected_response in response_message for expected_response in expected_responses]), (
"Expected links or serper not setup in response but got: " + response_message
)
assert any(
[expected_response in response_message for expected_response in expected_responses]
), f"Expected links: {expected_responses}. Actual response: {response_message}"
# ----------------------------------------------------------------------------------------------------
@pytest.mark.skipif(
os.getenv("SERPER_DEV_API_KEY") is None or os.getenv("OLOSTEP_API_KEY") is None,
reason="requires SERPER_DEV_API_KEY and OLOSTEP_API_KEY",
)
@pytest.mark.chatquality
@pytest.mark.django_db(transaction=True)
def test_chat_with_online_webpage_content(chat_client):
@@ -100,9 +99,9 @@ def test_chat_with_online_webpage_content(chat_client):
# Assert
expected_responses = ["185", "1871", "horse"]
assert response.status_code == 200
assert any([expected_response in response_message for expected_response in expected_responses]), (
"Expected links or serper not setup in response but got: " + response_message
)
assert any(
[expected_response in response_message for expected_response in expected_responses]
), f"Expected links: {expected_responses}. Actual response: {response_message}"
# ----------------------------------------------------------------------------------------------------