2.3.91 Satellite Open Design

Open Design

Handle: open-design
URL: http://localhost:34900

Open Design is a local-first design artifact workspace for prototypes, decks, dashboards, posters, mobile screens, and other visual outputs. It serves a web UI and daemon from one container, persists projects in .od, and can use BYOK cloud APIs or Harbor Ollama as the model provider.

Starting

harbor pull open-design
harbor up open-design --open

Open Design starts at http://localhost:34900. On first launch, open Settings and configure a model provider if you run the service by itself.

For a fully local Harbor model path, start it with Ollama:

harbor up open-design ollama --open

When Open Design starts with Ollama, Harbor configures the browser app for:

Provider: OpenAI-compatible
Base URL: http://127.0.0.1:11434/v1
API key: ollama
Model: llama3.2:1b

The 127.0.0.1:11434 URL is inside the Open Design container. Harbor runs a small loopback proxy in the same network namespace and forwards it to the ollama service, which keeps Open Design inside its upstream SSRF protections while still making the Harbor backend usable. Open Design talks to Ollama through Ollama's OpenAI-compatible /v1 API.

Configuration

Environment Variables

Following options can be set via harbor config:

HARBOR_OPEN_DESIGN_HOST_PORT             # Host port for the Open Design web UI and API
HARBOR_OPEN_DESIGN_IMAGE                 # Open Design runtime image
HARBOR_OPEN_DESIGN_VERSION               # Open Design image tag
HARBOR_OPEN_DESIGN_WORKSPACE             # Persistent data root
HARBOR_OPEN_DESIGN_ALLOWED_ORIGINS       # Comma-separated browser origins allowed to call /api
HARBOR_OPEN_DESIGN_MEM_LIMIT             # Compose memory limit
HARBOR_OPEN_DESIGN_NODE_OPTIONS          # Node.js runtime options
HARBOR_OPEN_DESIGN_API_PROTOCOL          # First-run provider protocol when started without an integration
HARBOR_OPEN_DESIGN_BASE_URL              # First-run provider base URL
HARBOR_OPEN_DESIGN_API_KEY               # First-run provider API key
HARBOR_OPEN_DESIGN_MODEL                 # First-run provider model
HARBOR_OPEN_DESIGN_PROVIDER_BASE_URL     # First-run provider preset URL
HARBOR_OPEN_DESIGN_FORCE_DEFAULTS        # Overwrite existing browser provider config on load
HARBOR_OPEN_DESIGN_OLLAMA_MODEL          # Model used by the Harbor Ollama integration
HARBOR_OPEN_DESIGN_LLAMACPP_MODEL        # Model id sent to the Harbor llama.cpp integration
HARBOR_OPEN_DESIGN_VLLM_MODEL            # Model id sent to the Harbor vLLM integration

Example:

harbor config set open-design.ollama.model llama3.2:1b
harbor config set open-design.force.defaults true
harbor up open-design ollama --open

Set open-design.force.defaults back to false after forcing a one-time browser reconfiguration.

Volumes

• ${HARBOR_OPEN_DESIGN_WORKSPACE}/data -> /app/.od - Open Design projects, SQLite database, app config, and runtime data.
• services/open-design/start.sh -> /usr/local/bin/harbor-open-design-start.sh - Harbor entrypoint that injects first-run provider defaults before starting the upstream daemon.

Harbor Integrations

Open Design can use Harbor Ollama as its local model provider:

harbor up open-design ollama
harbor up open-design llamacpp
harbor up open-design vllm

Available cross-service files:

Backend	Cross-file	Behavior
Ollama	compose.x.open-design.ollama.yml	Adds a loopback proxy at 127.0.0.1:11434 inside Open Design and preconfigures the app for HARBOR_OPEN_DESIGN_OLLAMA_MODEL through Ollama's OpenAI-compatible /v1 API
llama.cpp	compose.x.open-design.llamacpp.yml	Preconfigures Open Design for http://llamacpp:8080/v1 with HARBOR_OPEN_DESIGN_LLAMACPP_MODEL
vLLM	compose.x.open-design.vllm.yml	Preconfigures Open Design for http://vllm:8000/v1 with HARBOR_OPEN_DESIGN_VLLM_MODEL

HARBOR_OPEN_DESIGN_LLAMACPP_MODEL=auto asks Open Design's Harbor startup shim to read the first model advertised by llama.cpp's /v1/models endpoint and use that in the browser config.

Open Design's upstream container does not bundle local coding-agent CLIs such as Codex, Claude Code, Gemini CLI, or OpenCode. Use the BYOK provider path for Harbor-managed container runs.

If the browser already has older Open Design settings, either update Settings manually or force the Harbor defaults once:

harbor config set open-design.force.defaults true
harbor up open-design llamacpp --open
harbor config set open-design.force.defaults false

Security

The upstream deploy guide warns that Open Design's API is unauthenticated for non-browser clients. Harbor publishes the service on the configured host port for local use. If exposing it through a domain, public IP, or reverse proxy, set allowed origins explicitly:

harbor config set open-design.allowed.origins "https://od.example.com,http://203.0.113.10:34900"

Troubleshooting

docker logs harbor.open-design
docker logs harbor.open-design-ollama-proxy

If the UI opens but model calls fail with an upstream or provider error:

harbor config get open-design.ollama.model
harbor ollama list

Pull or choose a model that exists in Ollama, then restart:

harbor ollama pull llama3.2:1b
harbor up open-design ollama

If the browser already has different Open Design provider settings, either update Settings manually or force Harbor defaults once:

harbor config set open-design.force.defaults true
harbor up open-design ollama --open
harbor config set open-design.force.defaults false

Links

• Official Deployment Guide
• GitHub Repository