GonkaGate Qwen Code Setup

Connect an existing Qwen Code install to GonkaGate with the official installer. This guide uses the supported @gonkagate/qwen-code-setup path, so you do not have to hand-edit Qwen settings, export keys through shell profiles, or store secrets in a repository.

Quick setup with the official installer

Run the official installer:

Quick setup with the official installer

npx @gonkagate/qwen-code-setup

This tool configures an existing local qwen install. It does not install Qwen Code itself.

Use it if you want the safe default setup without hand-editing Qwen Code config:

keeps the secret out of repository-local files
preserves unrelated Qwen settings instead of rewriting whole files
checks GonkaGate /v1/models before writing config
writes the supported GonkaGate models into Qwen Code’s OpenAI-compatible provider catalog
verifies the durable local setup before returning you to plain qwen

Before you start

Qwen Code is already installed locally and available as qwen on PATH.
The audited Qwen Code baseline is @qwen-code/qwen-code 0.18.0.
Node >=22.14.0 is available for the npx installer.
You already have a GonkaGate API key in gp-... format.
The installer currently starts with qwen/qwen3-235b-a22b-instruct-2507-fp8.

Choose the right scope

The installer supports two scopes:

user: use GonkaGate on this machine in general.
project: make GonkaGate the default for the current repository.

Project scope keeps secrets out of repository files. The installer writes the GonkaGate provider catalog and GONKAGATE_API_KEY into the active user Qwen settings, then writes only activation settings into project Qwen settings.

One limit matters here: project modelProviders can hide user-managed providers in the audited Qwen Code baseline. If a trusted project already defines modelProviders, the installer blocks instead of writing a setup that would look correct but fail at runtime.

Non-interactive setup

Use this for scripts, automation, or repeatable local setup:

Non-interactive setup

npx @gonkagate/qwen-code-setup --scope user --yes

With a key from the environment:

With a key from the environment

GONKAGATE_API_KEY="$GONKAGATE_API_KEY" npx @gonkagate/qwen-code-setup --scope project --yes

With a key through stdin and JSON output:

With a key through stdin and JSON output

printf '%s' "$GONKAGATE_API_KEY" | npx @gonkagate/qwen-code-setup --api-key-stdin --scope project --yes --json

Use dry run when you want to inspect planned writes first:

Command

npx @gonkagate/qwen-code-setup --scope user --dry-run

Pass the key only through:

the hidden interactive prompt
GONKAGATE_API_KEY
--api-key-stdin

The installer intentionally does not accept plain --api-key, so secrets do not end up in shell history or process lists.

Verify the setup

If Qwen Code is already open, close it first. Then start a fresh session:

Verify the setup

qwen

Then:

confirm the active model is a GonkaGate model
switch models in Qwen Code if needed
send this prompt:

Reply with exactly: Qwen Code connected to GonkaGate

That confirms the provider, model selection, and request path before a longer coding session.

The installer verifies durable local settings by default. If you explicitly want a live model call during setup, use --verify-live. That can spend quota and depends on current provider and network availability, so it is not part of the default success check.

Where qwen-code-setup writes data

The installer keeps secrets out of the repository. It manages these Qwen Code surfaces:

user settings: ~/.qwen/settings.json
QWEN_HOME user settings: <QWEN_HOME>/settings.json
project activation settings: .qwen/settings.json
backups: ~/.gonkagate/qwen-code/backups/
install state: ~/.gonkagate/qwen-code/install-state.json

The managed Qwen settings use:

modelProviders.openai[] for GonkaGate models
security.auth.selectedType = "openai"
model.name for the selected GonkaGate model
settings.env.GONKAGATE_API_KEY for the durable key

The provider catalog includes all currently supported GonkaGate models for this setup flow:

qwen/qwen3-235b-a22b-instruct-2507-fp8
moonshotai/Kimi-K2.6
minimaxai/minimax-m2.7

Common first failures

If you see	What it usually means	What to do
`qwen_not_found`	Qwen Code is missing or not on `PATH`	Install `@qwen-code/qwen-code`, then rerun the installer
`qwen_version_unsupported`	The local Qwen Code version does not match the audited setup contract	Use the supported baseline or rerun the compatibility audit before changing setup assumptions
`secret_missing`	No safe GonkaGate key source was provided	Use the hidden prompt, `GONKAGATE_API_KEY`, or `--api-key-stdin`
`required_models_unavailable`	GonkaGate `/v1/models` did not return every required model for this key	Recheck the key and model access, then retry
`project_modelproviders_override`	Trusted project settings define `modelProviders` and would hide the user provider set	Remove the project override or use user scope
`secret_shadowed_by_process_env`	The current shell has a different `GONKAGATE_API_KEY`	Unset or align the current environment value before starting `qwen`
`secret_shadowed_by_project_env`	A trusted project `.qwen/.env` or `.env` overrides the managed user key	Remove or align the project env value
`live_verify_failed` after `--verify-live`	Local config can be correct while the live request fails for quota, model, or network	Check balance, quota, selected model, and provider availability before retrying live verification

Current limits

This guide configures an existing Qwen Code install. It does not install Qwen Code itself.
The audited Qwen Code baseline is @qwen-code/qwen-code 0.18.0.
The current transport target is OpenAI-compatible chat/completions through https://api.gonkagate.com/v1.
The installer writes all three supported models into modelProviders.openai[], then selects one through model.name.
The installer does not accept arbitrary custom base URLs.
The installer does not accept arbitrary raw model IDs.
The installer does not accept plain --api-key.
The installer does not mutate shell profiles or store secrets in repository-local files.
Default setup success is based on local verification, not a live model call.

Qwen Code

Quick setup with the official installer

Before you start

Choose the right scope

Non-interactive setup

Verify the setup

Where qwen-code-setup writes data

Common first failures

Current limits

Official references

See also

On this page