ModelStack

Agent teams

One model id.
A whole team.

A coordinator model, the specialists it can call, and your choice of who writes the reply. Called like any model.

account_treeagents / my-coding-agentCANVAS
Advises · coordinator writes the reply
Delegates · reply is sent as written

One coordinator, and the specialists it brings in.

Every team has exactly one coordinator. It answers routine requests on its own and consults a specialist when a request needs one, up to five rounds per reply.

One large coordinator node linked to three smaller specialist nodes. Two links return to the coordinator, one continues straight on to the right.
Advice comes back to the coordinator. A delegated reply goes straight through.

Six specialists to choose from

Each role can be added once, on any model in the catalog.

  • Senior Architect

    Architecture, design patterns, security and performance trade-offs.

  • Researcher

    APIs, framework docs and libraries you have not used before.

  • Coder

    Complex implementations, algorithms, tests and refactors.

  • Reviewer

    Security audits, bug detection and code-quality review.

  • UI/UX Consultant

    Interface design, user flows, accessibility and design systems.

  • Front-end Consultant

    Component architecture, state, performance and build tooling.

Decide who writes the reply.

Every specialist has a response mode. It changes what the user gets back, not how the coordinator asks for help.

Advise

Default

The coordinator writes the reply.

  1. Request
  2. Coordinator
  3. Senior Architect
  4. advisesCoordinator
  5. Reply

The specialist's answer goes back to the coordinator as advice. The coordinator combines it with everything else it knows and writes the final response.

Prompt
“Should we shard the orders table now, or wait?”
What comes back
The Senior Architect weighs in. The coordinator writes one answer with the trade-offs in the context of your codebase.

Delegate

The answer goes straight to the user.

  1. Request
  2. Coordinator
  3. Reviewer
  4. delegatesReply

When the specialist gives a final answer, it is sent to the user exactly as written. The coordinator hands off and adds nothing.

Prompt
“Review this diff for security issues.”
What comes back
The Reviewer's findings arrive exactly as it wrote them. Nothing is summarised or softened on the way back.
  • Advise is the default for every specialist.
  • Only specialists can delegate. The coordinator answers in its own words.
  • A delegating specialist that needs more context still asks through the coordinator first.

Build it on a canvas. Call it with one string.

Add a coordinator and specialists, configure each one where it sits, save. From then on the team id is a model id in your code.

1from openai import OpenAI
2 
3client = OpenAI(api_key="sk_...", base_url="https://api.modelstack.cc/v1")
4 
5stream = client.chat.completions.create(
6    model="my-coding-agent",  # the team id is the model
7    messages=[{"role": "user", "content": "Review this diff"}],
8    stream=True,
9)
tune

Configure each agent in place

Select a node and its settings open beside the canvas. Drag nodes to lay the team out the way you think about it.

  • Model, from the full catalog
  • Fallback model if the first is unavailable
  • Custom rules, in plain language
  • Response mode: Advise or Delegate
2request formats
OpenAI chat completions and Anthropic messages, on the same endpoint.
SSEstreaming
Streams like a single model. A delegated reply arrives as one chunk.
$0team surcharge
Each agent bills at its own model's rate for the tokens it uses.

Common questions.

How do I call an agent team?expand_more

Use the team's id as the model in any chat completions or messages request. The same API key, request format and streaming work exactly as they do for a single model.

What is the difference between Advise and Delegate?expand_more

Advise sends the specialist's answer back to the coordinator, which combines it with everything else it knows and writes the reply. Delegate sends the specialist's final answer to the user exactly as written. Advise is the default, and only specialists can delegate.

Does the coordinator always consult a specialist?expand_more

No. It answers routine requests on its own and consults a specialist when a request needs one, up to five consultation rounds per reply.

How is an agent team billed?expand_more

Each agent is charged at its own model's rate for the tokens it uses, so a reply costs the coordinator's tokens plus any specialist's tokens. There is no extra charge for using a team, and provider tiers apply as usual.

Which specialists can I add?expand_more

Six roles: Senior Architect, Researcher, Coder, Reviewer, UI/UX Consultant and Front-end Consultant. Each can be added once, with its own model, an optional fallback model and custom rules.

How many teams can I create?expand_more

It depends on your plan: Starter 3, Builder 10, Pro 10, Power unlimited, Max unlimited. Teams are private to your account and each one has an id you choose, which stays fixed after creation.

Ready to build your first team?

One coordinator, the specialists you pick, and one model id in your code.