Tribe AI

✨ Low code tool to rapidly build and coordinate multi-agent teams ✨

• Table of Contents
• What is Tribe?
• What are some use cases?
• Highlights
• How to get started
  • Generate Secret Keys
  • Deploy Tribe locally with Docker (simplest way)
  • Deploy Tribe on a remote server
• Guides and concepts
  • Sequential vs Hierarchical workflows
    • Sequential workflows
    • Hierarchical workflows
  • Skills
    • Create a Skill Using Skill Definitions
    • Writing a Custom Skill using LangChain
  • Retrieval Augmented Generation (RAG)
    • Customising embedding models
  • Guides
    • Creating Your First Hierarchical Team
    • Equipping Your Team Member with Skills
    • Creating Your First Sequential Team
    • Requiring Human Approval Before Skill Execution in Sequential Workflows
• Contribution
• Release Notes
• License

Have you heard the saying, 'Two minds are better than one'? That's true for agents too. Tribe leverages on the langgraph framework to let you customize and coordinate teams of agents easily. By splitting up tough tasks among agents that are good at different things, each one can focus on what it does best. This makes solving problems faster and better.

By teaming up, agents can take on more complex tasks. Here are a few examples of what they can do together:

• ⚽️ Footbal analysis: Imagine a team of agents where one scours the web for the latest Premier League news, and another analyzes the data to write insightful reports on each team's performance in the new season.
• 🏝️ Trip Planning: For planning your next vacation, one agent could recommend the best local eateries, while another finds the top-rated hotels for you. This team makes sure every part of your trip is covered.
• 👩‍💻 Customer Service: A customer service team where one agent handles IT issues, another manages complaints, and a third takes care of product inquiries. Each agent specializes in a different area, making the service faster and more efficient.

and many many more!

• Persistent conversations: Save and maintain chat histories, allowing you to continue conversations.
• Observability: Monitor and track your agents’ performance and outputs in real-time using LangSmith to ensure they operate efficiently.
• Tool Calling: Enable your agents to utilize external tools and APIs.
• Retrieval Augmented Generation: Enable your agents to reason with your internal knowledge base.
• Human-In-The-Loop: Enable human approval before tool calling.
• Easy Deployment: Deploy Tribe effortlessly using Docker.
• Multi-Tenancy: Manage and support multiple users and teams.

Before deploying it, make sure you change at least the values for:

• SECRET_KEY
• FIRST_SUPERUSER_PASSWORD
• POSTGRES_PASSWORD

You can (and should) pass these as environment variables from secrets.

Some environment variables in the .env file have a default value of changethis.

You have to change them with a secret key, to generate secret keys you can run the following command:

python -c "import secrets; print(secrets.token_urlsafe(32))"

Copy the content and use that as password / secret key. And run that again to generate another secure key.

Get up and started within minutes on your local machine.

Deploy Tribe on your remote server.

In a sequential workflow, your agents are arranged in an orderly sequence and execute tasks one after another. Each task can be dependent on the previous task. This is useful if you want to tasks to be completed one after another in a deterministic sequence.

Use this if:

• Your project has clear, step-by-step tasks.
• The outcome of one task influences the next.
• You prefer a straightforward and predictable execution order.
• You need to ensure tasks are performed in a specific order.

In a hierarchical workflow, your agents are organised into a team-like structure comprising of 'team leader', 'team members' and even other 'sub-team leaders'. The team leader breaks down the task into smaller tasks and delegate them to its team members. After the team members complete these tasks, their responses will be passed to the team leader who then chooses to return the response to the user or delegate more tasks.

Use this if:

• Your tasks are complex and multifaceted.
• You need specialized agents to handle different subtasks.
• Task delegation and re-evaluation are crucial for your workflow.
• You want flexibility in task management and adaptability to changes.

Skills are abilities that you can equip your agents with to interact with the world. For example, you can provide your agent with the skill to check the current weather condition or search the web for the latest news. By default, Tribe provides three skills:

• duckduckgo-search: Performs web searches.
• wikipedia: Searches Wikipedia for information.
• yahoo-finance: Retrieves information from Yahoo Finance News.

You will likely want to create custom skills, which can be done in two ways: by using function definitions for simple HTTP requests or by writing custom skills in the codebase.

If your skill involves performing an HTTP request to fetch or update data, using skill definitions is the simplest approach. In Tribe, start by navigating to the 'Skills' tab and clicking the 'Add Skill' button. You will then be prompted to provide the skill definition, which instructs your agent on how to execute the specific skill. This definition should be structured as follows:

{
  "url": "https://example.com",
  "method": "GET",
  "headers": {},
  "type": "function",
  "function": {
    "name": "Your skill name",
    "description": "Your skill description",
    "parameters": {
      "type": "object",
      "properties": {
        "param1": {
          "type": "integer",
          "description": "Description of the first parameter"
        },
        "param2": {
          "type": "string",
          "enum": ["option1"],
          "description": "Description of the second parameter"
        }
      },
      "required": ["param1", "param2"]
    }
  }
}

For more intricate tasks that extend beyond simple HTTP requests, LangChain allows you to develop more advanced tools. You can integrate these tools into Tribe by adding them to the managed_skills dictionary. For a practical example, refer to the demo calculator tool. To learn how to create a LangChain tool, please consult their documentation.

After creating a new tool, restart the application to ensure the tool is properly loaded into the database. Likewise, if you need to remove a tool, simply delete it from the managed_skills dictionary and restart the application to ensure it is removed from the database. Do note that tools created this way are available to all users in your application.

RAG is a technique for augmenting your agents' knowledge with additional data. Agents can reason about a wide range of topics, but their knowledge is limited to public data up to the point in time they were trained on. If you want your agents to reason about private data, Tribe allows you to upload your data and select which data to include in your agent’s knowledge base. This enables your agents to reason with the selected data and allows you to create different agents with specialized knowledge.

By default, Tribe uses BAAI/bge-small-en-v1.5, which is a light and fast English embedding model that is better than OpenAI Ada-002. If your documents are multilingual or require image embedding, you may want to use another embedding model. You can easily do this by changing DENSE_EMBEDDING_MODEL in your .env file:

# See the list of supported models: https://qdrant.github.io/fastembed/examples/Supported_Models/
DENSE_EMBEDDING_MODEL=BAAI/bge-small-en-v1.5 # Change this

Log into Tribe using the email and password you set during the installation step.

Navigate to the 'Teams' page and click on 'Add Team'. Enter a name for your team and click 'Save'.

Create two additional team members by dragging the handle of the Team Leader node.

Update the first team member as shown.

Update the second team member as shown.

Go to the 'Chat' tab and send a question to your team to see how they respond.

Congratulations! You’ve successfully built and communicated with your first multi-agent team on Tribe.

Your team member can do more by providing it with a set of skills. Add a skill to your Foodie.

Now, when you ask your Foodie a question, it will search the web for more up-to-date information!

Create a new team and select the 'Sequential' workflow.

Drag and drop to create another team member below 'Worker0'.

Update the first team member as shown. Provide the 'wikipedia' skill to this team member.

Update the second team member as shown.

Go to the 'Chat' tab and send a question to your team to see how they respond. Notice that the Researcher will use Wikipedia to do its research. Very cool!

You can require your team members to wait for your approval before executing their skills. Edit your Researcher and select 'Require approval'.

Now, before the Researcher executes its skills, it will ask you to approve or reject.

If you approve, the Researcher will continue to execute its skills.

Tribe is open sourced and welcome contributions from the community! Check out our contribution guide to get started.

Some ways to contribute:

• Report bugs and issues.
• Enhance our documentation.
• Suggest or contribute new features or enhancements.

Check the file release-notes.md.

Tribe is licensed under the terms of the MIT license.