VibeSpace uses BYOK (Bring Your Own Key): configure your own credentials and endpoint details. Model requests go directly from your Mac to the configured service — VibeSpace runs no relay, never sees your content, and takes no markup.
Supported providers
The catalog includes 28 connection options: 16 model providers, 9 aggregation and inference platforms, 2 cloud services, and 1 custom OpenAI-compatible endpoint option. The custom option can connect to services such as Ollama and LM Studio; these are examples of that option, not additional built-in providers.
Model providers (16)
| Provider | Notes |
|---|---|
| DeepSeek | DeepSeek language models |
| GLM | GLM models from Zhipu |
| Moonshot | Kimi models from Moonshot |
| MiniMax | MiniMax language models |
| Qwen | Qwen models from Alibaba |
| Doubao | Doubao models from ByteDance |
| OpenAI | OpenAI model API |
| Anthropic | Claude models |
| Gemini models | |
| StepFun | StepFun language models |
| Hunyuan | Tencent Hunyuan / TokenHub model service |
| Qianfan | Baidu Qianfan model service |
| Spark | iFLYTEK Spark; use an HTTP API Password and enter the model ID manually |
| xAI | Grok models |
| Mistral AI | Mistral model service |
| Cohere | Cohere Command models |
Aggregation and inference platforms (9)
| Provider | Notes |
|---|---|
| SiliconFlow | SiliconFlow model platform |
| Groq | Groq inference platform |
| Cerebras | Cerebras inference platform |
| Together AI | Open-model inference platform |
| Fireworks AI | Model inference platform |
| Hugging Face | Inference Providers aggregation service |
| Cloudflare Workers AI | Account-specific /ai/v1 Base URL; enter the model ID manually |
| NVIDIA NIM | NVIDIA hosted model inference |
| OpenRouter | Access models from multiple providers |
Cloud services (2)
| Provider | Notes |
|---|---|
| Azure OpenAI | Resource-specific /openai/v1 Base URL and deployment name |
| Amazon Bedrock | Regional Chat Completions endpoint and model ID; Bedrock API Key only |
Custom services (1)
| Provider | Notes |
|---|---|
| OpenAI Compatible | Custom Base URL and model ID for an OpenAI-compatible service, such as Ollama or LM Studio |
Setup
- Open Settings → LLM Providers and choose one of the 28 options;
- Enter your API Key (Spark uses an HTTP API Password);
- Fill in the Base URL when required. Azure OpenAI, Amazon Bedrock, Cloudflare Workers AI, and custom endpoints require their own endpoint URL;
- For Spark, Cloudflare Workers AI, and Amazon Bedrock, enter the model ID manually; for Azure OpenAI, enter the deployment name. Custom endpoints let you add model IDs manually;
- Click Test Connection. Where model discovery is supported, choose from the returned model list; then save the connection;
- Optional: override the model's context window limit if auto-detection is wrong.
You can configure multiple connections and choose one per task.
How your key is protected
- Keys are encrypted with AES-256-GCM and stored separately from everyday config;
- Decrypted only in memory while sending a request — never written to logs;
- Keys go to the provider endpoint and nowhere else — not to VibeSpace's website or servers.
Network proxy
If a provider endpoint requires a proxy, go to Settings → System → Network:
- Direct: default, connect straight to the provider;
- System: follow macOS proxy settings;
- Manual: enter a custom proxy address.
Usage tracking
Settings → Personal shows a token usage heatmap: day-by-day consumption to gauge cost and intensity per task.
Tip: context windows vary widely between models. For long tasks pick a large-window model and watch actual usage in the context panel.