1
00:00:00,760 --> 00:00:03,970
In this video, we ll talk about language
chain tools.

2
00:00:04,720 --> 00:00:08,810
Language chain tools are like specialized
apps for LLM.

3
00:00:09,700 --> 00:00:15,470
They are tiny code modules that allow it
to access information and services.

4
00:00:17,260 --> 00:00:24,090
These tools connect your LLM to search
engines, databases, APIs, and more,

5
00:00:24,480 --> 00:00:27,230
expanding its knowledge and capabilities.

6
00:00:28,120 --> 00:00:30,470
Think of them as building blocks for

7
00:00:30,480 --> 00:00:33,990
creating smarter, more versatile language models.

8
00:00:35,120 --> 00:00:38,130
So whether your AI needs real -time

9
00:00:38,140 --> 00:00:44,190
weather updates, wants to search
Wikipedia or even translate languages, it

10
00:00:44,200 --> 00:00:46,910
can use these tools to get the job done.

11
00:00:48,280 --> 00:00:52,330
Now, let s explore some interesting and

12
00:00:52,340 --> 00:00:52,970
useful tools.

13
00:00:53,920 --> 00:00:57,070
The first one we ll discuss is a search tool.

14
00:00:57,980 --> 00:01:04,210
There are several search tools available,
such as Google Search, DuckDuckGo, or

15
00:01:04,220 --> 00:01:11,230
Taveli AI, which is optimized for LLMs
and provides an API for AI developers.

16
00:01:12,540 --> 00:01:19,030
In this video, we ll focus on DuckDuckGo
search tool to avoid the hassle of

17
00:01:19,040 --> 00:01:22,590
setting up Google API keys or other components.

18
00:01:24,920 --> 00:01:28,010
First, let s install DuckDuckSearch.

19
00:01:35,850 --> 00:01:48,340
Let s import the necessary library from
langchain tools import DuckDuckGoSearchRun.

20
00:01:50,680 --> 00:01:58,870
I am creating an object of type
DuckDuckGoSearchRun, search equals, and I

21
00:01:58,880 --> 00:02:01,030
am calling the constructor of the class.

22
00:02:03,870 --> 00:02:06,420
I am running the tool by calling search

23
00:02:06,430 --> 00:02:10,000
.invoke with the prompt string as an argument.

24
00:02:10,730 --> 00:02:16,300
Output equals, search .invoke, and the

25
00:02:16,310 --> 00:02:20,810
prompt Where was Freddie Mercury born?

26
00:02:24,470 --> 00:02:25,780
And print output.

27
00:02:26,690 --> 00:02:34,020
I am running the code, and I ve got the result.

28
00:02:35,410 --> 00:02:38,700
It searched the web using this service

29
00:02:38,710 --> 00:02:42,080
and gave me this result.

30
00:02:42,710 --> 00:02:45,320
You can ignore this warning for now.

31
00:02:46,070 --> 00:02:49,960
It s about running DuckDuckGo in the
asynchronous mode.

32
00:02:51,730 --> 00:02:57,440
The tool has some attributes that can be
printed using the . notation.

33
00:02:58,310 --> 00:03:01,560
search .name will print the name of the tool.

34
00:03:03,130 --> 00:03:07,880
search .description will print its description.

35
00:03:10,630 --> 00:03:16,660
To get more additional information, such
as the link, use DuckDuckGo search results.

36
00:03:18,850 --> 00:03:29,530
I am importing the class from linkchain
.tools import DuckDuckGo search results,

37
00:03:30,860 --> 00:03:41,320
and I am creating an object, and output
equals, search .run, and the prompt.

38
00:03:41,750 --> 00:03:55,160
The prompt will be Freddie Mercury and
Quinn, and I am printing the output.

39
00:03:56,290 --> 00:03:57,520
I am running the code.

40
00:03:59,010 --> 00:04:02,100
As you can see, output is a string

41
00:04:02,110 --> 00:04:06,820
representation of a list of dictionaries
with the following keys.

42
00:04:20,630 --> 00:04:27,260
If you want to have much more control
over the search results, you can directly

43
00:04:27,270 --> 00:04:34,140
pass a custom DuckDuckGo search API
wrapper to DuckDuckGo search results.

44
00:04:35,130 --> 00:04:47,280
From linkchain .community .utilities
import DuckDuckGo search API wrapper, and

45
00:04:47,290 --> 00:04:54,040
I am creating a wrapper object,
wrapper equals DuckDuckGo search API

46
00:04:54,050 --> 00:05:02,720
wrapper of region equals, and the country
code, for example, de -de from

47
00:05:02,730 --> 00:05:13,790
Deutschland or Germany, max underline
results equals 3, and the safe search

48
00:05:13,800 --> 00:05:27,280
equals moderate, and the search object,
DuckDuckGo search results of API wrapper

49
00:05:27,290 --> 00:05:42,460
equals wrapper, and the source equals
news, output equals search .run, and

50
00:05:42,470 --> 00:05:45,780
Berlin will be the argument or the prompt.

51
00:05:46,890 --> 00:05:50,980
I am running the code and printing the

52
00:05:50,990 --> 00:05:59,850
output, and by the way, if you want to
get rid of the warnings, you can do

53
00:05:59,860 --> 00:06:09,570
import warnings and warnings
.filterWarnings of ignore as a string.

54
00:06:11,020 --> 00:06:16,830
I am running it, now the warnings, which
are not errors, will be ignored.

55
00:06:22,390 --> 00:06:26,020
Let us display the output in a more user
-friendly way.

56
00:06:26,770 --> 00:06:32,440
I will use regular expression to extract
the snippet, the link, and the title.

57
00:06:33,570 --> 00:06:39,120
I am importing the regular expression
module, import re.

58
00:06:40,550 --> 00:06:48,870
Next, I will define the pattern for the
regular expression, the pattern includes

59
00:06:48,880 --> 00:06:55,710
3 groups, each group corresponds to the
snippet, title, and link in the string,

60
00:06:56,660 --> 00:06:59,010
snippet, title, and link.

61
00:06:59,020 --> 00:07:03,890
Next, I will call the findAll function to

62
00:07:03,900 --> 00:07:12,610
find all matches of the pattern in the
string, matches equals re .findAll of

63
00:07:12,620 --> 00:07:25,980
pattern output and the re .thatAll, the
re .thatAll flag is used so that the

64
00:07:25,990 --> 00:07:30,000
.character matches any character
including a newline.

65
00:07:31,470 --> 00:07:39,260
Now, let us loop through the matches for
snippet, title, and link in matches,

66
00:07:39,510 --> 00:07:43,300
colon, and the for loop.

67
00:07:44,050 --> 00:07:47,400
Each match is a tuple containing the

68
00:07:47,410 --> 00:07:49,380
snippet, the title, and the link.

69
00:07:49,810 --> 00:07:52,480
I will unpack the tuple into variables.

70
00:07:52,490 --> 00:08:00,480
I am printing the snippet, the title, and
the link using an F string.

71
00:08:01,990 --> 00:08:05,840
I am also printing a separator for readability.

72
00:08:10,300 --> 00:08:11,130
I am running the code.

73
00:08:12,600 --> 00:08:15,850
Sorry, I have mistyped pattern and snippet.

74
00:08:18,560 --> 00:08:22,190
As you can see, it returned 3 results in

75
00:08:22,200 --> 00:08:23,770
German about Berlin.

76
00:08:23,780 --> 00:08:27,630
Let us go ahead and talk about another

77
00:08:27,640 --> 00:08:29,990
useful tool, the Wikipedia tool.

78
00:08:31,400 --> 00:08:35,470
I am installing it, pip install -q -wikipedia.

79
00:08:38,700 --> 00:08:41,690
I am importing the necessary modules.

80
00:08:42,700 --> 00:08:45,290
From the linkchain -community -tools, I

81
00:08:45,300 --> 00:08:50,650
am importing Wikipedia -Query -Run and
from the linkchain -community -utilities,

82
00:08:50,660 --> 00:08:54,630
I am importing the Wikipedia -API
-Wrapper class.

83
00:08:56,360 --> 00:08:58,830
Now, we initialize the tool.

84
00:08:59,380 --> 00:09:02,110
This is where we can configure it as we want.

85
00:09:02,880 --> 00:09:16,740
API -Wrapper equals Wikipedia -API
-Wrapper of top -k results equals 1 and

86
00:09:16,750 --> 00:09:21,420
the doc -content -chars -max equals 100.

87
00:09:24,500 --> 00:09:27,970
Now, I am creating an object of type

88
00:09:27,980 --> 00:09:29,190
Wikipedia -Query -Run.

89
00:09:30,350 --> 00:09:37,880
Wiki equals Wikipedia -Query -Run of API

90
00:09:37,890 --> 00:09:40,580
-Wrapper equals API -Wrapper.

91
00:09:41,890 --> 00:09:48,340
Let us use 500 instead of 100 and we can

92
00:09:48,350 --> 00:09:56,160
call this tool with a dictionary as
input, wiki .invoke and a dictionary,

93
00:09:58,180 --> 00:10:01,690
query, colon, lama -index.

94
00:10:05,030 --> 00:10:05,520
I am running.

95
00:10:07,670 --> 00:10:12,800
By the way, lama -index is a data
framework that helps build smart

96
00:10:12,810 --> 00:10:17,560
applications by connecting LLMs with your
specific information.

97
00:10:19,150 --> 00:10:26,130
Very well, I got some information about
the lama -index from Wikipedia.

98
00:10:26,860 --> 00:10:32,170
Note that the output is cut off at 500 characters.

99
00:10:33,080 --> 00:10:37,430
I am replacing 500 with 5000 to get the

100
00:10:37,440 --> 00:10:38,130
entire content.

101
00:10:42,680 --> 00:10:46,350
You can also call this tool with a string

102
00:10:46,360 --> 00:10:51,970
as an argument, wiki .invoke of Google -Gemini.

103
00:11:00,970 --> 00:11:03,740
That is it about the doc -doc -go and

104
00:11:03,750 --> 00:11:04,800
Wikipedia tools.

105
00:11:05,470 --> 00:11:06,560
Thank you, bye -bye.