Skip to content

2.3.98 Satellite AutoGen Studio

Jump to bottom
av edited this page Sep 25, 2026 · 1 revision

AutoGen Studio

Handle: autogenstudio
URL: http://localhost:34960

AutoGen Studio dashboard

AutoGen Studio is a low-code interface for prototyping agents and multi-agent workflows. Harbor builds the published autogenstudio Python package into an isolated container.

Starting

harbor build autogenstudio
harbor up autogenstudio --open

No account is required by the app. AutoGen Studio is a research prototype without authentication, so Harbor binds its port to 127.0.0.1 by default. Do not expose it to an untrusted network. It can execute agent tools and reach other Harbor services on harbor-network.

Configuration

The defaults are in services/autogenstudio/default.env. Set them with harbor config set and run harbor config update after changing defaults in a checkout:

HARBOR_AUTOGENSTUDIO_HOST_PORT=34960
HARBOR_AUTOGENSTUDIO_BIND_ADDRESS=127.0.0.1
HARBOR_AUTOGENSTUDIO_IMAGE="harbor/autogenstudio"
HARBOR_AUTOGENSTUDIO_VERSION="0.4.2.2"
HARBOR_AUTOGENSTUDIO_WORKSPACE="./services/autogenstudio/data"

The workspace is mounted at /data, where Studio stores its SQLite database and generated files. harbor down autogenstudio leaves the workspace intact. To pass additional environment variables, use harbor env autogenstudio <KEY> <VALUE>.

Local models

Start a Harbor Ollama model alongside Studio:

harbor up autogenstudio ollama
harbor ollama pull qwen3:4b

In Team Builder, select a team and switch Visual Builder to View JSON. Set the participant's model_client.config to a Harbor Ollama model, then click Save Changes:

{
  "model": "qwen3:4b",
  "api_key": "ollama",
  "base_url": "http://ollama:11434/v1",
  "model_info": {
    "vision": false,
    "function_calling": true,
    "json_output": false,
    "family": "unknown",
    "structured_output": true
  }
}

These are the fields inside model_client.config, not a replacement for the whole team JSON. Ollama does not check the API key. Match model_info capabilities to the selected model. The URL uses Docker's internal ollama hostname, not localhost, because Studio runs in its own container. See AutoGen's model-client guidance.

Troubleshooting

If the UI fails to start, inspect docker logs harbor.autogenstudio for the package or database error. If a local model cannot be reached, confirm Ollama is running with harbor ps and check that the model base URL uses http://ollama:11434/v1.

Links

Home | CLI Reference | Services | Adding New Service | Compatibility

Clone this wiki locally