AI Agent Tools

Our new Graphlit Agent Tools for Python enable easy interaction with agent frameworks such as CrewAI, allowing developers to easily integrate the Graphlit Platform into agentic workflows.

We have example Google Colab notebooks using CrewAI, which provide an example for analyzing the web marketing strategy of a company, and for structured data extraction of products from scraped web pages.

!pip install --upgrade
!pip install --upgrade
!pip install --upgrade
from crewai import Agent, Task, Crew, Process
from graphlit_tools import WebSearchTool, CrewAIConverter

web_search_tool = CrewAIConverter.from_tool(WebSearchTool(graphlit))

web_search_agent = Agent(
    role="Web Researcher",
    goal="Find the {company} website.",
    backstory="",
    verbose=True,
    allow_delegation=False,
    tools=[web_search_tool],
)

search_web_task = Task(
    description=(
        """Given company named {company}, search the web to find their home page.
        Return the root path for URLs, not individual web pages.
        For example return https://www.example.com, 
		not https://www.example.com/index.html"""
    ),
    expected_output="A single URL for the {company} home page",
    agent=web_search_agent,
)

crew = Crew(
    agents=[web_search_agent],
    tasks=[search_web_task],
    process=Process.sequential,
    planning=True,
    verbose=True,
)

result = await crew.kickoff_async(inputs={"company": company_name})
print(result)
from crewai import Agent, Task, Crew, Process
from graphlit_tools import WebSearchTool, CrewAIConverter

web_search_tool = CrewAIConverter.from_tool(WebSearchTool(graphlit))

web_search_agent = Agent(
    role="Web Researcher",
    goal="Find the {company} website.",
    backstory="",
    verbose=True,
    allow_delegation=False,
    tools=[web_search_tool],
)

search_web_task = Task(
    description=(
        """Given company named {company}, search the web to find their home page.
        Return the root path for URLs, not individual web pages.
        For example return https://www.example.com, 
		not https://www.example.com/index.html"""
    ),
    expected_output="A single URL for the {company} home page",
    agent=web_search_agent,
)

crew = Crew(
    agents=[web_search_agent],
    tasks=[search_web_task],
    process=Process.sequential,
    planning=True,
    verbose=True,
)

result = await crew.kickoff_async(inputs={"company": company_name})
print(result)
from crewai import Agent, Task, Crew, Process
from graphlit_tools import WebSearchTool, CrewAIConverter

web_search_tool = CrewAIConverter.from_tool(WebSearchTool(graphlit))

web_search_agent = Agent(
    role="Web Researcher",
    goal="Find the {company} website.",
    backstory="",
    verbose=True,
    allow_delegation=False,
    tools=[web_search_tool],
)

search_web_task = Task(
    description=(
        """Given company named {company}, search the web to find their home page.
        Return the root path for URLs, not individual web pages.
        For example return https://www.example.com, 
		not https://www.example.com/index.html"""
    ),
    expected_output="A single URL for the {company} home page",
    agent=web_search_agent,
)

crew = Crew(
    agents=[web_search_agent],
    tasks=[search_web_task],
    process=Process.sequential,
    planning=True,
    verbose=True,
)

result = await crew.kickoff_async(inputs={"company": company_name})
print(result)

Content Ingestion

SlackIngestTool: Ingests messages from Slack channel

Ingests messages from Slack channel into knowledge base.

Accepts Slack channel name.

Returns extracted Markdown text and metadata from messages.

DiscordIngestTool: Ingests messages from Discord channel

Ingests messages from Discord channel into knowledge base.

Accepts Discord channel name.

Returns extracted Markdown text and metadata from messages.

MicrosoftTeamsIngestTool: Ingests messages from Microsoft Teams channel

Ingests messages from Microsoft Teams channel into knowledge base.

Returns extracted Markdown text and metadata from messages.

LinearIssueIngestTool: Ingests issues from Linear project

Ingests issues from Linear project into knowledge base.

Accepts Linear project name.

Returns extracted Markdown text and metadata from issues.

JiraIssueIngestTool: Ingests issues from Atlassian Jira

Ingests issues from Atlassian Jira into knowledge base.

Accepts Atlassian Jira server URL and project name.

Returns extracted Markdown text and metadata from issues.

GitHubIssueIngestTool: Ingests issues from GitHub repository

Ingests issues from GitHub repository into knowledge base.

Accepts GitHub repository owner and repository name.

Returns extracted Markdown text and metadata from issues.

GoogleEmailIngestTool: Ingests emails from Google Email account

Ingests emails from Google Email account into knowledge base.

Returns extracted Markdown text and metadata from emails.

MicrosoftEmailIngestTool: Ingests emails from Microsoft Email account

Ingests emails from Microsoft Email account into knowledge base.

Returns extracted Markdown text and metadata from emails.

RSSIngestTool: Ingests posts from RSS feed

Ingests posts from RSS feed into knowledge base.

For podcast RSS feeds, audio will be transcribed and ingested into knowledge base.

Returns extracted or transcribed Markdown text and metadata from RSS posts.

NotionIngestTool: Ingests pages from Notion database

Ingests pages from Notion database into knowledge base.

Returns extracted Markdown text and metadata from Notion pages.

RedditIngestTool: Ingest posts from Reddit subreddit

Ingests posts from Reddit subreddit into knowledge base.

Returns extracted Markdown text and metadata from Reddit posts.

WebMapTool: Map web site by URL

Accepts web page URL as string.

Enumerates the web pages at or beneath the provided URL using web sitemap.

Returns list of mapped URIs from web site.

WebSearchTool: Search web by text

Accepts search query text as string.

Performs web search based on search query.

Returns Markdown text and metadata extracted from web pages.

WebCrawlTool: Crawl web site from URL

Crawls web pages from web site into knowledge base.

Returns Markdown text and metadata extracted from web pages.

WebScrapeTool: Scrape web page from URL

Scrapes web page into knowledge base.

Returns Markdown text and metadata extracted from web page.

LocalIngestTool: Ingest from Local File

Accepts file path to local file.

Returns extracted Markdown text and metadata from content.

Can ingest individual Word documents, PDFs, audio recordings, videos, images, or any other unstructured data.

URLIngestTool: Ingest from URL

Accepts URL to cloud-hosted file.

Returns extracted Markdown text and metadata from content.

Can ingest individual Word documents, PDFs, audio recordings, videos, images, or any other unstructured data.

RAG

RAG

PromptTool: Prompt LLM using RAG pipeline

Accepts user prompt as string.

Prompts LLM with relevant content and returns completion from RAG pipeline. Returns Markdown text from LLM completion.

Uses vector embeddings and similarity search to retrieve relevant content from knowledge base.

Can search through web pages, PDFs, audio transcripts, and other unstructured data.

Data Retrieval

OrganizationRetrievalTool: Retrieve organizations

Accepts search text as string.

Retrieves organizations based on similarity search from knowledge base.

Returns metadata from organizations relevant to the search text.

PersonRetrievalTool: Retrieve persons

Accepts search text as string.

Retrieves persons based on similarity search from knowledge base.

Returns metadata from persons relevant to the search text.

ContentRetrievalTool: Retrieve content from knowledge base

Accepts search text as string.

Optionally accepts a list of content types (i.e. FILE, PAGE, EMAIL, ISSUE, MESSAGE) for filtering the result set.

Retrieves contents based on similarity search from knowledge base.

Returns extracted Markdown text and metadata from contents relevant to the search text.

Can search through web pages, PDFs, audio transcripts, Slack messages, emails, or any unstructured data ingested into the knowledge base.

Content Generation

GenerateChaptersTool: Generate chapters from transcript

Accepts transcript as string.

Returns chapters as text.

GenerateKeywordsTool: Generate keywords from text

Accepts text as string.

Optionally accepts the count of keywords to be generated.

Returns keywords as text.

GenerateQuestionsTool: Generate followup questions from text

Accepts text as string.

Optionally accepts the count of followup questions to be generated.

Returns followup questions as text.

GenerateSocialMediaPostsTool: Generate social media posts from text

Accepts text as string.

Optionally accepts the count of social media posts to be generated.

Returns social media posts as text.

GenerateHeadlinesTool: Generate headlines from text

Accepts text as string.

Optionally accepts the count of headlines to be generated.

Returns headlines as text.

GenerateBulletsTool: Generate bullet points from text

Accepts text as string.

Optionally accepts the count of bullet points to be generated.

Returns bullet points as text.

GenerateSummaryTool: Generate summary from text

Accepts text as string.

Optionally accepts text prompt to be provided to LLM for text summarization.

Returns summary as text.

Image Description

DescribeWebPageTool: Screenshot and describe web page

Screenshots web page from URL and describes web page with vision LLM.   

Returns Markdown description of screenshot and extracted Markdown text from image.

DescribeImageTool: Describe image by URL

Accepts image URL as string.

Prompts vision LLM and returns completion.

Returns Markdown text from LLM completion.

Data Extraction

ExtractTextTool: Extracts JSON data from text using LLM

Accepts text to be scraped, and JSON schema of Pydantic model to be extracted into.

JSON schema needs be of type 'object' and include 'properties' and 'required' fields.

Returns extracted JSON from text.

ExtractWebPageTool: Extracts JSON data from ingested web page using LLM

Accepts URL to be scraped, and JSON schema of Pydantic model to be extracted into.

JSON schema needs be of type 'object' and include 'properties' and 'required' fields.

Returns extracted JSON from web page.

ExtractURLTool: Extracts JSON data from ingested file using LLM

Accepts URL to be ingested, and JSON schema of Pydantic model to be extracted into.

JSON schema needs be of type 'object' and include 'properties' and 'required' fields.

Returns extracted JSON from file.