feat: add Mixlayer as an LLM provider

[cathrynlavery]

What

Add Mixlayer as an OpenAI-compatible LLM provider: docs page, overview row, sidebar nav, core provider class, GUI model picker, llm-info catalog, and openai-adapters routing.

Why

Mixlayer serves Qwen, DeepSeek, and GLM models at https://mixlayer.ai/v1 with bearer auth and zero data retention. Continue already has the same pattern for other OpenAI-compatible hosts (MiniMax, Tensorix, Mimo).

Details

Default model: z-ai/glm-5.3. API base in code is https://mixlayer.ai/v1/ (trailing slash, same as the other OpenAI-compatible providers). Docs use https://mixlayer.ai/v1. Env/secret name: MIXLAYER_API_KEY.

Model	ID	Context	Input	Output	Tools	Reasoning	Input $/M	Output $/M
GLM 5.3	z-ai/glm-5.3	262144	text	text	yes	always on (Low/High/Max; none rejected)	1.15	3.90
Qwen3.8 27B	qwen/qwen3.8-27b	262144	text, image	text	yes	optional (default on; none disables)	0.25	1.95
Qwen3.6 35B A3B	qwen/qwen3.6-35b-a3b	131072	text, image	text	yes	optional (default on; none disables)	0.25	1.30
DeepSeek V4.1 Flash	deepseek/deepseek-v4.1-flash	262144	text	text	yes	optional (default High; none disables; integer 1-100 accepted)	0.30	1.20
Qwen3.5 9B	qwen/qwen3.5-9b	131072	text, image	text	yes	optional (default on; none disables)	0.10	0.40
Qwen3.5 4B (Free)	qwen/qwen3.5-4b-free	131072	text, image	text	yes	optional (default on; none disables)	0	0

Quantization: GLM 5.3 and DeepSeek V4.1 Flash are fp8; the Qwen models are bf16. All six support structured outputs.

Testing

curl -sS -o /dev/null -w "%{http_code}\n" https://mixlayer.ai/v1/models
# 401

node ./scripts/build-packages.js
# all packages built successfully

cd core && npm run vitest -- llm/llms/OpenAI-compatible.vitest.ts -t mixlayer
# 7 passed, 211 skipped

cd core && npm test -- --testPathPattern=llm/index.test --testNamePattern=mixlayer --no-coverage
# 6 passed, 186 skipped (context length for all six model IDs)

cd core && npx tsc -p ./tsconfig.json --noEmit --pretty false --incremental false
# exit 0

npx prettier --write <touched ts/json files>
# unchanged

Did not run the full core suite, GUI typecheck, or the docs dev server.

Notes

• Max output tokens are not published. maxCompletionTokens is optional in @continuedev/llm-info, so it is omitted. Context length is the published value.
• Chat Completions does not support tool_choice. Continue only sends it when toolChoice is set, so it is documented rather than stripped.
• extensions/vscode/config_schema.json was not updated. That JSON-config enum is already missing merged providers (minimax, mimo, nous). YAML provider is a free string.
• Cache-read prices exist in the platform API but are not on the pricing page, so they are omitted.
• Canonical base URL used here is https://mixlayer.ai/v1 (docs). The homepage and models.dev also list https://models.mixlayer.ai/v1; both serve the same API today.
• Continue requires a CLA comment on the opened PR: I have read the CLA Document and I hereby sign the CLA. This change was not pushed and no PR was opened.

[github-actions]

All contributors have signed the CLA ✍️ ✅
_{Posted by the CLA Assistant Lite bot.}

[cathrynlavery]

I have read the CLA Document and I hereby sign the CLA