Cherry Studio
Cherry Studio is an open-source desktop LLM client for Windows / macOS / Linux with built-in multi-provider support.
π‘ One-click config: create a key on the API Keys page, then use Config Guide β Cherry Studio to copy the API Host and key with the real value filled in.
Download
cherry-ai.com β pick the installer for your OS.
Steps
- Open Cherry Studio β Settings β Model Services
- Select OpenAI (or create a new custom provider) and open its settings
- Fill in:
| Field | Value |
|---|---|
| API Host | https://api.tokensmart.ai/v1 |
| API Key | pk_live_xxxxxxxxxxxxxxxx |
- Click Check to test the connection
- Switch to the Models tab β Add, enter the model IDs you want (
gpt-5.4,claude-sonnet-4-6,gemini-2.5-pro, β¦) - Start a new chat β your models show up in the model picker
Using Claude
Cherry Studio also ships an Anthropic provider. To use the Anthropic protocol:
- API Host:
https://api.tokensmart.ai(no /v1) - API Key:
pk_live_xxxxxxxxxxxxxxxx
Tokensmart tips
- Run both providers side-by-side: point both the OpenAI and Anthropic providers at Tokensmart and you can A/B the same question against GPT and Claude inside one client β no key swap needed.
- Use the Anthropic path for prompt caching: Cherry Studio's Anthropic provider forwards
cache_controlβ perfect for long-running knowledge-base or RAG sessions where you can get the system prompt down to 1/10 of list price. - One key per Assistant: if you use Cherry Studio's multi-Assistant feature (writing, translation, code, support, β¦) give each one its own Tokensmart key and filter API Logs by key to see per-persona cost.
- Skip auto model discovery: "Auto-fetch models" hits
/v1/models, which only lists models your key can reach. Manually adding the 2β3 models you actually use keeps the picker clean.
Troubleshooting
- Connection failed β check whether
/v1is included (OpenAI protocol needs it; Anthropic protocol does not) - Models not found β auto-discovery calls
/v1/models, which only returns models your key has access to. Adding manually is more reliable - Streaming glitches β upgrade to the latest Cherry Studio; older builds had SSE parsing bugs