Power of Teams: Multi-Agent Collaboration 🤝

Why use one giant model when you can build a team of small, highly specialized local agents?

Single agents hit a ceiling fast. Their context windows get crowded, their reasoning gets muddy, and fixing them becomes a debugging nightmare.

This is especially true for local 1-7B parameter models. A 3B model is too small to be a master researcher, writer, database admin, and coder all at once.

But what if you split those roles among specialized agents and let them work as a Team? 🤝

A Team coordinates multiple specialized agents (and sub-teams) to tackle complex tasks. A leader agent coordinates the request and delegates sub-tasks to members based on their designated expertise.

🗺️ Visualizing Team Structure

Here's how a team divides labor under a leader:

Rendering diagram...

🛠️ Code: Building a Local Team

Here's how to build a translation team consisting of English, Chinese, and German specialists running locally:

1from kern.team import Team
2from kern.agent import Agent
3from kern.models.openai import OpenAIChat
4
5# Connect to local Ollama server 💻
6local_model = OpenAIChat(id="llama3.2:3b", base_url="http://localhost:11434/v1")
7
8translator_team = Team(
9    name="Global Translation Agency",
10    members=[
11        Agent(
12            name="English Translator",
13            model=local_model,
14            role="English localization expert",
15            instructions="Translate inputs accurately into natural English.",
16        ),
17        Agent(
18            name="Chinese Translator",
19            model=local_model,
20            role="Chinese localization expert",
21            instructions="Translate inputs accurately into natural Chinese.",
22        ),
23        Agent(
24            name="German Translator",
25            model=local_model,
26            role="German localization expert",
27            instructions="Translate inputs accurately into natural German.",
28        ),
29    ]
30)
31
32# Run the team! The coordinator will delegate tasks to the appropriate members. 🚀
33translator_team.print_response("Translate the welcome message: 'Welcome to Kern, your local agent engine!' into German and Chinese", stream=True)

⚖️ Why use Teams?

Benefit	How it helps (especially for Local Models!)
Domain Specialization 🧠	Each agent focuses on a narrow prompt, making 3B models behave like domain experts.
Parallel Execution ⚡	Independent sub-tasks run concurrently, speeding up total runtimes.
Easy Maintenance 🛠️	If the German translation breaks, you only edit the German translator agent, not the whole pipeline.
Smarter Context Control 📁	Individual context histories stay clean instead of cluttering a single giant chat context.

⚙️ Team Modes: How Leaders Collaborate

Kern offers three collaboration topologies via TeamMode. You choose how the team leader delegates tasks to members:

TeamMode.coordinate (Default): The leader acts as a project manager, dynamically chatting with members, collecting reports, and synthesizing the final answer.
TeamMode.broadcast: The leader sends the user query to all members simultaneously, gathers their outputs, and aggregates them. Great for comparison or multi-language translations!
TeamMode.route: The leader acts as a router, forwarding the query to the single most relevant agent who then responds directly.

1from kern.team import Team, TeamMode
2
3team = Team(
4    name="Quick Research Team",
5    members=[...],
6    mode=TeamMode.broadcast,  # Broadcast query to all members 📢
7)

📚 Guides & Next Steps

wrench

Build Teams 🛠️

Define custom members, modes, and coordination settings.

user-robot

Run Teams 🚀

Execute teams, process results, and stream responses.

bug

Debug Teams 🐛

Track intermediate delegation steps and execution timelines.