Graphlit's new Agent Tools for Python library enables easy interaction with agent frameworks such as CrewAI, allowing developers to easily integrate the power of the Graphlit Platform with their agentic workflows.
We have written 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.
As described in Anthropic's article "Building effective agents":
At Anthropic, we categorize all these variations as agentic systems, but draw an important architectural distinction between workflows and agents:
Workflows are systems where LLMs and tools are orchestrated through predefined code paths.
Agents, on the other hand, are systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks.
Our Agent Tool Library can be used with both Agentic Workflows and Agents, orchestrated by any LLM, from Anthropic Sonnet 3.5 to OpenAI GPT-4o to open-source models like Meta Llama 3.x.
Agent Tools provide knowledge, memory, text/image analysis and content generation - which can be used with Agents of any framework.
We provide a directory of the available Graphlit Agent Tools at the end of this tutorial.
Content Ingestion
RAG
Data Retrieval
Content Generation
Image Description
Data Extraction
In this tutorial, we show how to use the Agent Tools Library with CrewAI, but we will be offering adapters for other agent frameworks, such as LangGraph or Autogen in the future.
Get started
Before you begin, ensure you have the following:
- Python 3.x installed on your system.
- An active account on the Graphlit Platform with access to the API settings dashboard.
To install CrewAI:
pip install --upgrade crewai
pip install --upgrade
And then to install the Graphlit Agent Tools with CrewAI:
Initialize Graphlit client
The Graphlit Client supports environment variables to be set for authentication and configuration:
- GRAPHLIT_ENVIRONMENT_ID: Your environment ID.
- GRAPHLIT_ORGANIZATION_ID: Your organization ID.
- GRAPHLIT_JWT_SECRET: Your JWT secret for signing the JWT token.
Alternately, you can pass these values with the constructor of the Graphlit client.
You can find these values in the API settings dashboard on the Graphlit Platform.
For example, to use Graphlit in a Google Colab notebook, you need to assign these properties as Colab secrets: GRAPHLIT_ORGANIZATION_ID, GRAPHLIT_ENVIRONMENT_ID and GRAPHLIT_JWT_SECRET.
import os
from google.colab import userdata
from graphlit import Graphlit
os.environ['GRAPHLIT_ORGANIZATION_ID'] = userdata.get('GRAPHLIT_ORGANIZATION_ID')
os.environ['GRAPHLIT_ENVIRONMENT_ID'] = userdata.get('GRAPHLIT_ENVIRONMENT_ID')
os.environ['GRAPHLIT_JWT_SECRET'] = userdata.get('GRAPHLIT_JWT_SECRET')
graphlit = Graphlit()
Use Agent Tools in CrewAI
Once you have configured the Graphlit client, as shown below, you will pass the client to the tool constructor. The Graphlit client provides the authenticated integration to the Graphlit Platform API.
For use in CrewAI, you will need to convert the tool to the CrewAI tool schema with the CrewAIConverter.from_tool() function.
web_search_tool = CrewAIConverter.from_tool(WebSearchTool(graphlit)
Let's walkthrough how to integrate the Graphlit WebSearchTool, WebMapTool and DescribeWebPage tool in this CrewAI example for analyzing the web marketing strategy of a company.
The Colab notebook asks for the name of an automaker to be analyzed, ex: Land Rover.
Also, it creates instances of the Graphlit agent tools to be used by the CrewAI agents, passing the Graphlit client instance.
import os
import dotenv
from crewai import Agent, Crew, Process, Task
from langchain_openai import ChatOpenAI
from graphlit_tools import WebSearchTool, WebMapTool, DescribeWebPageTool, CrewAIConverter
from datetime import datetime
company_name = input('Enter the automaker company name to be analyzed: ')
web_search_tool = CrewAIConverter.from_tool(WebSearchTool(graphlit)
web_map_tool = CrewAIConverter.from_tool(WebMapTool(graphlit)
describe_web_page_tool = CrewAIConverter.from_tool(DescribeWebPageTool(graphlit)
Next, we will define the CrewAI agents, which will be joined into a crew. We are using the OpenAI GPT-4o model with these agents.
You can see how we assign the instance of the WebSearchTool to the 'Web Researcher' agent, the WebMapTool to the 'Web Mapping Agent', and the DescribeWebPageTool to the 'Web Analyst Agent'.
llm = ChatOpenAI(model="gpt-4o")
web_search_agent = Agent(
role="Web Researcher",
goal="Find the {company} website.",
backstory="",
verbose=True,
llm=llm,
allow_delegation=False,
tools=[web_search_tool],
)
web_map_agent = Agent(
role="Web Mapping Agent",
goal="Enumerate all the web page URLs for the provided web site.",
backstory="",
verbose=True,
llm=llm,
allow_delegation=False,
tools=[web_map_tool],
)
web_page_analyst_agent = Agent(
role="Web Analyst Agent",
goal="Visually analyze the {company} web pages, and describe the branding, overall layout and marketing approach. Also extract the company name and any products you find.",
backstory="You work for a major automotive manufacturer, and are doing competitive analysis on other automakers websites.",
verbose=True,
llm=llm,
allow_delegation=False,
tools=[describe_web_page_tool],
)
editor_agent = Agent(
role="Marketing Editor Agent",
goal="Write marketing strategy reports given provided web page analyses.",
backstory="You work for a major automotive manufacturer, and are doing competitive analysis on other automakers websites.",
verbose=True,
llm=llm,
allow_delegation=False
)
Next we will define the tasks to be executed by the agents.
The descriptions give instructions to the agents to perform the tasks, and specify the expected textual output from the tasks.
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:
),
expected_output="A single URL for the {company} home page",
agent=web_search_agent,
)
fetch_web_pages_task = Task(
description=(
"""Fetch the URLs at or beneath the given home page for further analysis.
Filter the resulting URLs to locate pages which appear to be about automobile models and specifications.
Select one most relevant page per automobile model.
"""
),
expected_output="A list of web page URLs, maximum 5",
agent=web_map_agent,
context=[search_web_task],
)
analyze_web_page_task = Task(
description=(
"""Analyze the provided web pages from the {company} website from a marketing perspective.
Execute task once for each provided web page.
Keep as much of the detail from each web page in your final analysis.
Do *not* pass a prompt to the provided tool, just skip the 'prompt' argument so the tool uses its default analysis prompt.
"""
),
expected_output="A thorough analysis of the web pages from a marketing perspective.",
agent=web_page_analyst_agent,
context=[fetch_web_pages_task],
)
writer_task = Task(
description=(
"""Write a thorough analysis of the {company} web marketing strategy given the detailed analyses from the provided {company} web pages.
Keep useful details from the web page analysis in your final summary.
"""
),
expected_output="A thorough and well-structured summary of the {company} web marketing strategy.",
agent=editor_agent,
context=[analyze_web_page_task],
)
Finally, we will kickoff the crew and print the result, while passing the 'company name' to the crew as an input.
crew = Crew(
agents=[web_search_agent, web_map_agent, web_page_analyst_agent, editor_agent],
tasks=[search_web_task, fetch_web_pages_task, analyze_web_page_task, writer_task],
process=Process.sequential,
planning=True,
verbose=True,
)
result = await crew.kickoff_async(inputs={"company": company_name})
print("Website Summary Process Completed:")
print(result)
Crew Output Walkthrough
In the crew output, you can see it planning the execution, and first calling the 'Web Researcher' agent to find the Land Rover home page. It uses the WebSearchTool to find the official website, and selects the best URL.
Enter the automaker company name to be analyzed: Land Rover
[2024-12-07 22:15:17][INFO]: Planning the crew execution
# Agent: Web Researcher
## Task: Given company named Land Rover, search the web to find their home page.
Return the root path for URLs, not individual web pages.
For example return https:
# Agent: Web Researcher
## Thought: I should perform a web search to find the official website of Land Rover.
## Using tool: Graphlit web search tool
## Tool Input:
"{\"search\": \"Land Rover official site\", \"search_limit\": 5}"
## Tool Output:
URL: https:
Title: Land Rover | Explore Luxury SUVs and 4x4 Vehicles
Jaguar Land Rover Limited is constantly seeking ways to improve the specification, design and production of its vehicles, parts, options and/or accessories and alterations take place continually, and we reserve the right to make changes without notice. The information, specification, engines and colors on this website are based on European specifications and may vary from market to market and are subject to change without notice. This is a very dynamic situation, and as a result imagery used within the website at present may not fully reflect current specifications for features, options, trim and color schemes. PLEASE NOTE
Choose from Range Rover, Defender or Discovery
© 2023 Jaguar Land Rover North America, LLC
Important note on imagery & specifications. The global shortages of semiconductors and other materials are currently affecting vehicle build specifications, option availability, and build timings.
URL: https:
Title: Defender | 4x4 Vehicles | Embrace the Impossible - Land Rover USA
Find the Land Rover vehicle that’s perfect for you with our latest special offers. BUILD YOUR OWN RANGE ROVER BUILD YOUR OWN DEFENDER BUILD YOUR OWN DISCOVERY SEARCH INVENTORY LAND ROVER EXPERIENCE --------------------- Discover a range of exhilarating driving experiences, guided tours and adventure travel with some of the world’s most capable vehicles. OWNERSHIP ACCESSORIES PLUG-IN ELECTRIC HYBRID (PHEV) SERVICE AND MAINTENANCE CONTACT LAND ROVER OWNERS STORIES -------------- Land Rover vehicle owners view things through a different lens. LAND ROVER ELECTRIC LAND ROVER ELECTRIC Jaguar Land Rover Limited is constantly seeking ways to improve the specification, design and production of its vehicles, parts, options and/or accessories and alterations take place continually, and we reserve the right to make changes without notice.
URL: https:
Title: Land Rover | Explore Luxury SUVs and 4x4 Vehicles
Jaguar Land Rover Limited is constantly seeking ways to improve the specification, design and production of its vehicles, parts and accessories and alterations take place continually, and we reserve the right to change without notice. ... The figures provided are as a result of official manufacturer's tests in accordance with EU legislation
URL: https:
Title: Home :: JLR Careers
We're evolving the world of modern luxury. Creating vehicles that are era-defining across distinct brands with profound cultural influence worldwide. Range Rover. Discovery. Defender. Jaguar. We're embracing our modernist design philosophy and delivering a more sustainable planet. To do this takes diverse skills and a collective purpose.
URL: https:
Title: Explore Land Rover Vehicle Models | Land Rover USA
Jaguar Land Rover Limited is constantly seeking ways to improve the specification, design and production of its vehicles, parts, options and/or accessories and alterations take place continually, and we reserve the right to make changes without notice. Some features may vary between optional and standard for different model year vehicles.
# Agent: Web Researcher
## Final Answer:
https:
Next, it uses the 'Web Mapping Agent' to enumerate the web pages (via sitemap) on the selected URL.
# Agent: Web Mapping Agent
## Task: Fetch the URLs at or beneath the given home page for further analysis.
Filter the resulting URLs to locate pages which appear to be about automobile models and specifications.
Select one most relevant page per automobile model.
1. The Web Mapping Agent will begin this task by accessing the Graphlit web map tool. 2. The agent will input the URL obtained from Task Number 1 into the 'url' argument. 3. The Web Mapping Agent will initiate the mapping process to enumerate the web pages at or beneath the provided URL. 4. The tool will scan the site and compile a list of all mapped URIs. 5. The Web Mapping Agent will filter through the resulting URLs to identify pages specifically related to automobile models and specifications. 6. The agent will select the most relevant page for each automobile model identified, ensuring a maximum of 5 URLs are documented. 7. The agent will compile the list of filtered URLs and prepare them as the output for this task.
# Agent: Web Mapping Agent
## Thought: I need to use the Graphlit web map tool to enumerate all the web page URLs at or beneath the given home page URL, "https://www.landrover.com". This will allow me to identify and filter pages related to automobile models and specifications.
## Using tool: Graphlit web map tool
## Tool Input:
"{\"url\": \"https://www.landrover.com\"}"
## Tool Output:
https:
https:
https:
https:
https:
https:
[truncated for example]
https:
https:
https:
https:
https:
https:
https:
# Agent: Web Mapping Agent
## Final Answer:
1. https:
2. https:
3. https:
4. https:
5. https:
Then, using the 'Web Analyst Agent', and the DescribeWebPageTool, the Crew will screenshot each mapped web page, and extract text about the Land Rover model and its specifications.
# Agent: Web Analyst Agent
## Task: Analyze the provided web pages from the Land Rover website from a marketing perspective.
Execute task once for each provided web page.
Keep as much of the detail from each web page in your final analysis.
Do *not* pass a prompt to the provided tool, just skip the 'prompt' argument so the tool uses its default analysis prompt.
1. The Web Analyst Agent will receive the list of URLs obtained from Task Number 2. 2. The agent will iterate through each URL one at a time. 3. For each URL, the Web Analyst Agent will access the Graphlit screenshot web page tool. 4. The agent will input the current web page's URL into the 'url' argument. 5. The agent will execute the tool to capture a screenshot of the webpage and extract details without passing a custom prompt, allowing for default analysis. 6. The tool will generate a description and extract relevant Markdown text from the screenshot. 7. The Web Analyst Agent will compile the analyses, focusing on branding, layout, marketing strategy, detailing the company name, and information about the products. 8. The agent will finalize a thorough analysis for each page and keep it organized for the next task.
# Agent: Web Analyst Agent
## Thought: I will start by analyzing the first URL provided to me, which is the page for the Land Rover Defender models and specifications.
## Using tool: Graphlit screenshot web page tool
## Tool Input:
"{\"url\": \"https://www.landrover.com/defender/defender/models-and-specifications.html\"}"
## Tool Output:
## Textual Content Analysis
### Extracted Text
**Specifications**
**Choose Your Bodystyle**
- Defender 90
- Defender 110
- Defender 130
**Choose Your Model**
- Defender S
- Key Features: 19" Style 6010 wheels, LED Headlights, Resist and Resolve Seats, 3D Surround Camera
- Defender X-Dynamic SE
- Key Features: 20" Style 5095 wheels, Satin Dark Grey wheels, LED headlights with signature DRL, Resist Seats, Meridian™ Sound System
- Defender X-Dynamic HSE
- Key Features: 20" Style 5095 wheels, Satin Dark Grey wheels, Matrix LED headlights with signature DRL, Windsor Leather Seats, Sliding panoramic roof
- Defender X
- Key Features: 22" Style 5098 wheels, Satin Dark Grey wheels, Electronic Air Suspension, Signature Interior Upgrade, ClearSight interior rear view mirror
**Standard Features**
**Legal and Technical Information**
- "*View Defender 90 Range*"
- Legal disclaimers regarding fuel economy, emissions, and technical specifications.
### Analysis
#### Key Points and Technical Terms
- **Bodystyle Options**: The webpage offers three bodystyle options for the Land Rover Defender: 90, 110, and 130. These numbers typically refer to the wheelbase length, which affects the vehicle's size and capacity.
- **Model Variants**: The models listed (S, X-Dynamic SE, X-Dynamic HSE, X) indicate different trim levels, each with unique features and specifications.
- **Key Features**: Each model is associated with specific features, such as wheel styles, lighting technology (LED, Matrix LED), seating materials (Resist, Windsor Leather), and additional functionalities (panoramic roof, ClearSight mirror).
#### Named Entities
- **Land Rover Defender**: The primary subject of the webpage, a well-known SUV model.
- **Meridian™ Sound System**: A branded audio system, indicating a focus on premium sound quality.
- **ClearSight**: A technology feature for enhanced rear visibility.
### Visual Elements
#### Imagery
- The images of the vehicles are central to the page, providing a visual representation of each model and bodystyle. This helps potential buyers visualize the differences between options.
#### Color Scheme and Branding
- The color scheme is predominantly neutral, with shades of grey and white, which aligns with the luxury and ruggedness associated with the Land Rover brand.
- The Land Rover logo is present, reinforcing brand identity.
### Readability and Organization
- **Layout**: The content is organized into clear sections, making it easy for users to navigate between bodystyle and model options.
- **Text Presentation**: Key features are listed in bullet points, enhancing readability and allowing for quick comparison between models.
### Communication Effectiveness
- The webpage effectively communicates the variety of options available for the Land Rover Defender, catering to different customer preferences and needs.
- The combination of textual descriptions and visual elements provides a comprehensive overview, aiding potential buyers in making informed decisions.
### Overall Assessment
The webpage successfully fulfills its promotional role by clearly presenting the different configurations of the Land Rover Defender. It targets potential buyers by highlighting key features and options, using both textual and visual elements to enhance understanding and engagement. The organization and clarity of the content facilitate a smooth user experience, likely increasing the likelihood of customer interaction and conversion
Finally, the 'Marketing Editor Agent' takes the extracted text from each web page, and publishes the final marketing report as the Crew response.
# Agent: Marketing Editor Agent
## Task: Write a thorough analysis of the Land Rover web marketing strategy given the detailed analyses from the provided Land Rover web pages.
Keep useful details from the web page analysis in your final summary.
1. The Marketing Editor Agent will collect all detailed analyses prepared by the Web Analyst Agent from Task Number 3. 2. The agent will review the analyses, pulling key insights and useful details regarding Land Rover's marketing strategy such as branding elements, target audiences, product positioning, and promotional tactics. 3. The Marketing Editor Agent will structure the summary logically, starting with an introduction to Land Rover's brand and concluding with overall marketing strategies observed across the analyzed web pages. 4. The agent will ensure that the summary is articulate, succinct, and well-supported by evidence from the page analyses. 5. The final summary will be prepared and formatted neatly for presentation, ensuring that it aligns with professional marketing analysis documentation standards.
# Agent: Marketing Editor Agent
## Final Answer:
### Land Rover Web Marketing Strategy Analysis
#### Introduction to Land Rover's Brand
Land Rover is a renowned automotive manufacturer known for its luxury and rugged vehicles. Its web marketing strategy is centered around showcasing its vehicles' premium quality, customization options, and advanced technology, appealing to affluent consumers who value both luxury and performance.
#### Detailed Analysis of Land Rover Web Pages
1. **Land Rover Defender Models and Specifications**
- **Branding and Layout:** The Land Rover Defender webpage effectively communicates the brand's rugged yet luxurious image. It offers three bodystyle options (90, 110, 130) and various model variants (S, X-Dynamic SE, X-Dynamic HSE, X), organized in a clean layout that enhances user experience. Neutral color schemes are used to maintain the brand's identity.
- **Marketing Approach:** The focus is on customization and luxury, highlighting key features like wheel styles, lighting technology, and interior materials. Consistent use of the Land Rover logo and branding elements reinforces brand identity.
- **Products Found:** Defender models - S, X-Dynamic SE, X-Dynamic HSE, X.
2. **Land Rover Discovery Sport Models and Specifications**
- **Branding and Layout:** The Discovery Sport webpage is minimalistic, with a focus on model differentiation. The use of bullet points for feature listings aids in easy comparison. A professional look is maintained with a white and black color scheme.
- **Marketing Approach:** The page targets family-friendly, versatile SUV consumers by emphasizing advanced technology and comfort features like Meridian™ Sound System and LED headlights. The "Build Your Own" option encourages user engagement and personalization.
- **Products Found:** Discovery Sport models - S, Dynamic SE, Dynamic HSE.
3. **Land Rover Range Rover Evoque Models and Specifications**
- **Branding and Layout:** The Evoque webpage is organized with sections for different models, complemented by high-quality images. This design approach supports the brand's luxury and technology focus.
- **Marketing Approach:** The emphasis is on appealing to urban, style-conscious consumers, highlighting luxury and technology features such as Pivi Pro and the Meridian™ Sound System.
- **Products Found:** Range Rover Evoque models - S, Dynamic SE, Autobiography.
4. **Land Rover Range Rover Sport Models and Specifications**
- **Branding and Layout:** The Range Rover Sport page features a straightforward layout for model comparison, consistent with the luxury branding of Range Rover.
- **Marketing Approach:** High performance and luxury are the focal points, with advanced features and customization options highlighted. The "Build Your Own" feature is key to engaging consumers.
- **Products Found:** Range Rover Sport models - Sport, S, SE, Dynamic SE.
5. **Land Rover Range Rover Velar Models and Specifications**
- **Branding and Layout:** The Velar page maintains a clean, professional look, focusing on model differentiation with high-quality images and a minimalist color scheme to enhance luxury appeal.
- **Marketing Approach:** Targeting luxury SUV buyers, the webpage emphasizes features like panoramic roofs and advanced sound systems, encouraging exploration and model comparison.
- **Products Found:** Range Rover Velar models - S, Dynamic SE, Autobiography.
#### Overall Marketing Strategies
Land Rover's website employs a consistent branding strategy focused on luxury, customization, and advanced technology. This approach is designed to appeal to a target market that values premium automotive experiences. Each webpage is meticulously organized, utilizing a blend of visuals and detailed descriptions to engage potential buyers and facilitate informed decision-making. Interactive elements like "Build Your Own" enhance user engagement and personalization, aligning with Land Rover's premium market positioning. This strategy effectively reinforces the brand's identity and supports its objectives of attracting and retaining a discerning customer base.
Website Summary Process Completed:
### Land Rover Web Marketing Strategy Analysis
#### Introduction to Land Rover's Brand
Land Rover is a renowned automotive manufacturer known for its luxury and rugged vehicles. Its web marketing strategy is centered around showcasing its vehicles' premium quality, customization options, and advanced technology, appealing to affluent consumers who value both luxury and performance.
#### Detailed Analysis of Land Rover Web Pages
1. **Land Rover Defender Models and Specifications**
- **Branding and Layout:** The Land Rover Defender webpage effectively communicates the brand's rugged yet luxurious image. It offers three bodystyle options (90, 110, 130) and various model variants (S, X-Dynamic SE, X-Dynamic HSE, X), organized in a clean layout that enhances user experience. Neutral color schemes are used to maintain the brand's identity.
- **Marketing Approach:** The focus is on customization and luxury, highlighting key features like wheel styles, lighting technology, and interior materials. Consistent use of the Land Rover logo and branding elements reinforces brand identity.
- **Products Found:** Defender models - S, X-Dynamic SE, X-Dynamic HSE, X.
2. **Land Rover Discovery Sport Models and Specifications**
- **Branding and Layout:** The Discovery Sport webpage is minimalistic, with a focus on model differentiation. The use of bullet points for feature listings aids in easy comparison. A professional look is maintained with a white and black color scheme.
- **Marketing Approach:** The page targets family-friendly, versatile SUV consumers by emphasizing advanced technology and comfort features like Meridian™ Sound System and LED headlights. The "Build Your Own" option encourages user engagement and personalization.
- **Products Found:** Discovery Sport models - S, Dynamic SE, Dynamic HSE.
3. **Land Rover Range Rover Evoque Models and Specifications**
- **Branding and Layout:** The Evoque webpage is organized with sections for different models, complemented by high-quality images. This design approach supports the brand's luxury and technology focus.
- **Marketing Approach:** The emphasis is on appealing to urban, style-conscious consumers, highlighting luxury and technology features such as Pivi Pro and the Meridian™ Sound System.
- **Products Found:** Range Rover Evoque models - S, Dynamic SE, Autobiography.
4. **Land Rover Range Rover Sport Models and Specifications**
- **Branding and Layout:** The Range Rover Sport page features a straightforward layout for model comparison, consistent with the luxury branding of Range Rover.
- **Marketing Approach:** High performance and luxury are the focal points, with advanced features and customization options highlighted. The "Build Your Own" feature is key to engaging consumers.
- **Products Found:** Range Rover Sport models - Sport, S, SE, Dynamic SE.
5. **Land Rover Range Rover Velar Models and Specifications**
- **Branding and Layout:** The Velar page maintains a clean, professional look, focusing on model differentiation with high-quality images and a minimalist color scheme to enhance luxury appeal.
- **Marketing Approach:** Targeting luxury SUV buyers, the webpage emphasizes features like panoramic roofs and advanced sound systems, encouraging exploration and model comparison.
- **Products Found:** Range Rover Velar models - S, Dynamic SE, Autobiography.
#### Overall Marketing Strategies
Land Rover's website employs a consistent branding strategy focused on luxury, customization, and advanced technology. This approach is designed to appeal to a target market that values premium automotive experiences. Each webpage is meticulously organized, utilizing a blend of visuals and detailed descriptions to engage potential buyers and facilitate informed decision-making. Interactive elements like "Build Your Own" enhance user engagement and personalization, aligning with Land Rover's premium market positioning. This strategy effectively reinforces the brand's identity and supports its objectives of attracting and retaining a discerning customer base.
In summary, we have shown how to integrate Graphlit Agent Tools with CrewAI, to offer integrations with web search, website mapping, and vision-based LLMs for screenshot text extraction.
Agent Tools Directory
Graphlit offers Agent Tools that cover a variety of use cases, such as:
Content Ingestion
RAG
Data Retrieval
Content Generation
Image Description
Data Extraction
Content Ingestion
URLIngestTool: Graphlit URL ingest tool
Ingests content from URL.
Returns extracted Markdown text and metadata from content.
Can ingest individual Word documents, PDFs, audio recordings, videos, images, or any other unstructured data.
LocalIngestTool: Graphlit local file ingest tool
Ingests content from 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.
WebScrapeTool: Graphlit web scrape tool
Scrapes web page into knowledge base.
Returns Markdown text and metadata extracted from web page.
WebCrawlTool: Graphlit web crawl tool
Crawls web pages from web site into knowledge base.
Returns Markdown text and metadata extracted from web pages.
WebSearchTool: Graphlit web search tool
Accepts search query text as string.
Performs web search based on search query.
Returns Markdown text and metadata extracted from web pages.
WebMapTool: Graphlit web map tool
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.
RedditIngestTool: Graphlit Reddit ingest tool
Ingests posts from Reddit subreddit into knowledge base.
Returns extracted Markdown text and metadata from Reddit posts.
NotionIngestTool: Graphlit Notion ingest tool
Ingests pages from Notion database into knowledge base.
Returns extracted Markdown text and metadata from Notion pages.
RSSIngestTool: Graphlit RSS ingest tool
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.
MicrosoftEmailIngestTool: Graphlit Microsoft Email ingest tool
Ingests emails from Microsoft Email account into knowledge base.
Returns extracted Markdown text and metadata from emails.
GoogleEmailIngestTool: Graphlit Google Email ingest tool
Ingests emails from Google Email account into knowledge base.
Returns extracted Markdown text and metadata from emails.
GitHubIssueIngestTool: Graphlit GitHub Issue ingest tool
Ingests issues from GitHub repository into knowledge base.
Accepts GitHub repository owner and repository name.
Returns extracted Markdown text and metadata from issues.
JiraIssueIngestTool: Graphlit Jira ingest tool
Ingests issues from Atlassian Jira into knowledge base.
Accepts Atlassian Jira server URL and project name.
Returns extracted Markdown text and metadata from issues.
LinearIssueIngestTool: Graphlit Linear ingest tool
Ingests issues from Linear project into knowledge base.
Accepts Linear project name.
Returns extracted Markdown text and metadata from issues.
MicrosoftTeamsIngestTool: Graphlit Microsoft Teams ingest tool
Ingests messages from Microsoft Teams channel into knowledge base.
Returns extracted Markdown text and metadata from messages.
DiscordIngestTool: Graphlit Discord ingest tool
Ingests messages from Discord channel into knowledge base.
Accepts Discord channel name.
Returns extracted Markdown text and metadata from messages.
SlackIngestTool: Graphlit Slack ingest tool
Ingests messages from Slack channel into knowledge base.
Accepts Slack channel name.
Returns extracted Markdown text and metadata from messages.
RAG
PromptTool: Graphlit RAG prompt tool
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
ContentRetrievalTool: Graphlit content retrieval tool
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.
PersonRetrievalTool: Graphlit person retrieval tool
Accepts search text as string.
Retrieves persons based on similarity search from knowledge base.
Returns metadata from persons relevant to the search text.
OrganizationRetrievalTool: Graphlit organization retrieval tool
Accepts search text as string.
Retrieves organizations based on similarity search from knowledge base.
Returns metadata from organizations relevant to the search text.
Image Description
DescribeImageTool: Graphlit image description tool
Accepts image URL as string.
Prompts vision LLM and returns completion. Returns Markdown text from LLM completion.
DescribeWebPageTool: Graphlit screenshot web page tool
Screenshots web page from URL and describes web page with vision LLM.
Returns Markdown description of screenshot and extracted Markdown text from image.
Content Generation
GenerateSummaryTool: Graphlit summary generation tool
Accepts text as string.
Optionally accepts text prompt to be provided to LLM for text summarization.
Returns summary as text.
GenerateBulletsTool: Graphlit bullet points generation tool
Accepts text as string.
Optionally accepts the count of bullet points to be generated.
Returns bullet points as text.
GenerateHeadlinesTool: Graphlit headlines generation tool
Accepts text as string.
Optionally accepts the count of headlines to be generated.
Returns headlines as text.
GenerateSocialMediaPostsTool: : Graphlit social media posts generation tool
Accepts text as string.
Optionally accepts the count of social media posts to be generated.
Returns social media posts as text.
GenerateQuestionsTool: Graphlit followup questions generation tool
Accepts text as string.
Optionally accepts the count of followup questions to be generated.
Returns followup questions as text.
GenerateKeywordsTool: Graphlit keywords generation tool
Accepts text as string.
Optionally accepts the count of keywords to be generated.
Returns keywords as text.
GenerateChaptersTool: Graphlit transcript chapters generation tool
Accepts transcript as string.
Returns time-stamped chapters as text.
Data Extraction
ExtractURLTool: Graphlit JSON URL data extraction tool
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.
ExtractWebPageTool: Graphlit JSON web page data extraction tool
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.
ExtractTextTool: Graphlit JSON text data extraction tool
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.
Summary
Please email any questions on this article or the Graphlit Platform to questions@graphlit.com.
For more information, you can read our Graphlit Documentation, visit our marketing site, or join our Discord community.
