Cumbersome: LLM API Client for iPhone and Mac

Cumbersome connects from your device to OpenAI, Anthropic, Google AI Studio, Z.ai, OpenRouter, Vercel AI Gateway, and compatible APIs. You can also reach a local server such as LM Studio or Ollama. Your keys stay in Apple Keychain, and Folding Sky never proxies the request.

Pick the exact model instead of accepting an automatic router. Message Details shows which provider and model answered, token and cache counts for the response, cumulative API tokens across the conversation, response timing, and a redacted copy of the request parameters. When a provider supports a setting Cumbersome doesn't expose, you can add it as a custom JSON parameter.

Annotations attach your note to an exact passage in a response, so a correction or constraint stays beside the text it belongs to. Face/Off sends one prompt through the selected model three times, then makes a fourth call to judge the answers. Mind/Meld uses three independent tracks and synthesizes their strongest ideas when a difficult prompt benefits from several approaches.

These controls cost what they actually use. A normal request makes one provider call. Face/Off and Mind/Meld make several calls, and the app tells you what is happening. You choose whether the extra comparison is worth the tokens.

Question Cards gives the model a structured way to ask up to three clarifying questions before answering. Cumbersome presents choices, rankings, sliders, dates, or a confirmation, then returns your answers to the model as a tool result. You can also skip the cards and let the model continue with reasonable assumptions.