Reef AI

Reef AI

AI on your own terms

0 ratings
Free

Rating summary

Details

  • Released
  • Updated
  • June 12, 2026
  • July 9, 2026

Features

Reef AI screenshot #1 for iPhone
Reef AI screenshot #2 for iPhone
Reef AI screenshot #3 for iPhone
Reef AI screenshot #4 for iPhone
Reef AI screenshot #5 for iPhone
iphone
ipad
🖼️Get Icon
Icons↘︎

About

Reef is an on-device AI harness for iPhone and iPad. Connect a local or hosted model, add tools and files when you need them, and keep chats and data on your device. You choose the LLM and the credentials. Reef does not include a subscription or relay your traffic. MODELS OpenAI, Anthropic, Mistral, Groq, OpenRouter, Perplexity, and OpenAI-compatible servers—Ollama, LM Studio, vLLM, LocalAI, Jan, and others. Multiple profiles with per-discussion settings. API keys stored in the iOS keychain. TOOLS Four types, same thread: • Platform — calendar, contacts, photos, health, location, web, workspaces, memory banks, and more. Runs on your phone; you grant each permission. • Endpoint — HTTP APIs you define. • Lens — structured reasoning with on-device signed records. • Dynamic — tools the model can register during a session. FILES & MEMORY Thread-scoped workspaces for PDFs, code, images, and documents. Named memory banks across sessions. Knowledge graph with a 3D explorer, built offline from your content. VOICE & AGENTS Composer dictation and Talk mode (hands-free with native device voices; additional TTS providers coming). Switch models mid-thread, coordinate several agents with @mentions, or run parallel discussions. PLUGINS Install skills, tools, prompts, and presets from Git or an archive—without waiting for an app update. LOCAL-FIRST Chats, embeddings, workspace files, and settings stay on device. Direct connection to the endpoint you configure. Full export and import. Offline queue and background streaming. Sensitive capabilities require explicit OS permission.
Show more

What's New in Reef AI

1.2.0

July 9, 2026

Reef 1.2 makes profiles lighter, faster, and easier to work with. Skills now load on demand, so profiles can carry richer instructions without inflating every prompt. Token budgeting is more accurate across prompts, tools, files, images, and model routes. Tool handling, memory, and multi-agent flows have also been tightened so agentic sessions feel more predictable. This release also includes reliability fixes, refreshed documentation, and cleanup across the app experience.

More