Vane

Run a self-hosted AI answer engine with TokenMix as the OpenAI provider.

Who It Is For

Use it if you want a self-hosted Perplexity-style answering engine. It combines search results with model responses and is useful for research, information gathering, and web Q&A.

Prepare TokenMix Values

• Provider / type: OpenAI Compatible or OpenAI
• Base URL: https://api.tokenmix.ai/v1
• API Key: TokenMix API key
• Model: a TokenMix model ID, for example <your-model-short-id>

What Each Field Means

• Provider / type decides which API protocol the app will use. Choose OpenAI Compatible, OpenAI, or Custom OpenAI only when it lets you set a Base URL.
• Base URL is the actual API endpoint. Use https://api.tokenmix.ai/v1, not the default OpenAI endpoint, and do not omit /v1.
• API Key is the key created in TokenMix. Treat it as a credential and do not put it in public repos, screenshots, examples, or chat logs.
• Model is a TokenMix model ID, not the app name. Avoid extra spaces and do not mix upstream model names with TokenMix display IDs.
• Embedding is only needed for document search, knowledge base indexing, and similarity search. For the first chat test, you can leave embedding unset.

Install Or Open The App

1. Start it with Docker or Node following the README.
2. Prepare a search service such as SearXNG.
3. Open Vane in your browser and go to settings.
4. Configure the model provider first, then the search source.

Choose A Deployment Option

• Easiest local option: Docker Bundle, including Vane and SearXNG.
• If you already have SearXNG: use Docker Slim and set SEARXNG_API_URL.
• One-click hosting: Sealos, RepoCloud, ClawCloud, and Hostinger.
• Developers: Node Local. After deployment, change the OpenAI provider Base URL to TokenMix in the setup screen.

Configure TokenMix

1. Add an OpenAI provider.
2. Set API Key to your TokenMix API key.
3. Set Base URL to https://api.tokenmix.ai/v1.
4. Set Chat Model to a TokenMix chat model ID.
5. Set Embedding Model to a TokenMix embedding model ID.
6. If the model list does not load, enter the model ID manually.

First Test

Search for “what is an OpenAI compatible API base URL” and answer briefly with sources.

Run the test once. If it succeeds, make one small real workflow or document test next. Do not connect production data before this basic call works.

What To Do After The First Test

• Confirm the search service returns results before judging answer quality.
• Configure both chat and embedding models, otherwise summaries or relevance may be unstable.
• If citations are missing, check the search backend before changing models.
• Use it for public web research first; do not immediately search internal sensitive data.

Troubleshooting

• 401 / unauthorized: wrong key, extra spaces, or an expired key.
• model not found: wrong model ID, or your account cannot access the model.
• Requests still go to OpenAI: Base URL was not changed to TokenMix, or the wrong provider was selected.
• Output is short or interrupted: send less context at once or use a larger-context model.

Checklist Before Real Use

• Test with a test key, test project, sample document, or manual trigger before using production data.
• Check app logs, browser Network, terminal output, or TokenMix request logs to confirm requests go to https://api.tokenmix.ai/v1.
• Make sure failures show understandable messages and do not expose API keys, upstream internals, or local paths.
• If the app can send email, update CRM, write databases, or run commands, start in draft-only or read-only mode.
• After it works, set team permissions, budget limits, and usage rules instead of sharing one high-privilege key.

Beginner Safety

Test with a temporary project, sample document, or manual trigger first. After normal responses work, connect real repositories, knowledge bases, or production automations. For file deletion, command execution, email sending, database changes, or CRM actions, ask for a plan first instead of direct execution.