2.3.99 Satellite Qwen Code

Qwen Code

Handle: qwencode
URL: http://localhost:34970

Qwen Code is a coding agent with a terminal interface and a browser Web Shell. Harbor runs the official Qwen Code image as a daemon with a persistent project workspace and user configuration.

Starting

harbor pull qwencode
harbor up --no-defaults qwencode
harbor open qwencode

The daemon binds inside its container on port 4170. Because the container listener is not loopback-only, Qwen Code requires a bearer token even if Docker publishes the port only on the host's loopback interface. With no configured token, Qwen Code generates one at startup; view it with docker logs --tail 30 harbor.qwencode. Treat that token as a password. For a stable token, set harbor config set qwencode.server.token <long-random-token> before startup. Harbor stores this value in its local .env file, so protect that file and never commit it. A newly created Harbor bridge network binds published ports to loopback on Harbor versions with that default; an existing or LAN-enabled network may publish this daemon on all interfaces. Do not expose the agent to an untrusted network.

Ollama integration

For a local model, pull a tool-capable model and start both services:

harbor pull qwen3.5:4b
harbor up --no-defaults qwencode ollama

On startup, Harbor adds the configured Ollama model to ~/.qwen/settings.json if it is not already listed, selects OpenAI-compatible authentication only if the user has not chosen another provider, and keeps other Qwen settings intact. The Ollama endpoint is http://ollama:11434/v1; sk-ollama is a placeholder API key because local Ollama does not require one. Change the default model with harbor config set qwencode.ollama.model <installed-model> and restart Qwen Code. An existing model entry with the same ID is left untouched; edit services/qwencode/config/settings.json if it points elsewhere. Model quality determines tool-use reliability.

The terminal CLI is available inside the running service:

harbor exec qwencode qwen --version
harbor exec qwencode qwen -p 'Summarize this workspace'

Qwen Code can read and modify files under the configured workspace. Mount only a project you intend the agent to access, and review tool approvals before allowing edits.

Configuration

Harbor setting	Default	Purpose
qwencode.host.port	34970	Host port for the Web Shell and daemon API
qwencode.image	ghcr.io/qwenlm/qwen-code	Official image repository
qwencode.version	0.24.5	Official image tag
qwencode.config	./services/qwencode/config	Persistent Qwen user settings and sessions
qwencode.workspace	./services/qwencode/workspace	Project directory mounted at /workspace
qwencode.server.token	empty	Optional stable daemon bearer token
qwencode.ollama.model	qwen3.5:4b	Default model seeded when Ollama runs alongside Qwen Code

Use harbor config get <setting> and harbor config set <setting> <value> to inspect or change these. The host config and workspace directories are bind-mounted into the container, so they survive container replacement. The official image runs as root and may create root-owned files there; host-side edits can require elevated permissions. services/qwencode/override.env can pass additional upstream environment variables through harbor env qwencode.

Troubleshooting

• harbor up waits for the daemon port to open. If the Web Shell asks for a token, read the startup token with docker logs --tail 30 harbor.qwencode or set a stable token and restart.
• If the model is missing, run harbor exec ollama ollama list, check qwencode.ollama.model, and start both services together so Harbor applies the integration overlay.
• If a prompt fails, check harbor logs qwencode and confirm the Ollama model responds independently before investigating Qwen Code.

Links

• Official documentation
• Qwen Code repository
• Daemon security and Web Shell