flowdown.ai

What It Is

FlowDown is a native AI workspace for iPhone, iPad, and Mac, positioned as “AI That Works Even Without Us.” It is not a single-model service, but a local-first multi-model client: it can connect to any OpenAI-compatible API, use Pollinations free models, Apple Intelligence, local MLX, or self-hosted endpoints, and sync across Apple devices via iCloud.

Core Capabilities

At the model layer, FlowDown supports both chat completions and responses APIs, with configurable endpoints, model IDs, headers, body fields, context length, temperature, reasoning/thinking parameters, and more. Feature-wise, it covers Markdown rendering, code highlighting, file attachments, vision, audio, web search, calendar, location, URL previews, memory, MCP toolchains, and Apple Shortcuts. It feels more like an AI control panel within the Apple ecosystem, suited to unifying multiple providers and local models in one interface.

Privacy and Integrations

Its biggest selling point is privacy: according to the documentation, conversations and API keys are stored locally on the device by default, with optional syncing through the user’s private iCloud/CloudKit. FlowDown says it does not collect data and connects directly to AI providers. The project is open source under AGPL-3.0, making it easier to audit. That said, users should still be careful: misconfigured custom endpoints, headers, or body fields may create data leakage or billing risks.

Pricing

FlowDown is distributed via the App Store. In the U.S. store, pricing decreases over time in tiers, and the basic app plan is scheduled to become free starting June 10, 2026. Core chat, model configuration, and local storage are included; future personalization features may be charged separately. The built-in free inference service and Pollinations models are suitable for testing, but the documentation clearly notes rate, regional, and availability limits. For long-term use, bringing your own API key or running local models is still recommended.

Pros, Cons, and Who It’s For

Pros include native performance, a privacy-focused architecture, broad multi-model compatibility, and strong automation capabilities via Shortcuts and MCP. Cons include Apple-platform exclusivity, relatively complex external model configuration, local inference support being described only for MLX with no direct GGUF/Core ML support, and no guarantee of stability for free models. It is best suited to heavy AI users, developers, self-hosted model users, and privacy-conscious Apple users. If you only want something that works out of the box, the official ChatGPT app or a simpler client may be less hassle.

Access from China

The crawled text does not provide details on availability or payment support in mainland China. Purchases depend on the App Store, with regional pricing determined by the store; model access depends on the configured third-party provider or local model. If cloud APIs are unstable, options include using a self-hosted OpenAI-compatible gateway, local MLX, or alternative tools such as Cherry Studio, Open WebUI, LM Studio, and Jan.