Llamatik vs Decubed

Llamatik — On-Device AI Chatbot Llamatik brings the power of lightweight open-source AI models directly to your iPhone or iPad. All text generation runs fully offline, on your device, without sending your conversations to servers. It’s fast, private, and designed to showcase what the Llamatik library can do on iOS. With Llamatik, you can explore model-based text generation, summarize content, brainstorm ideas, or simply chat — all powered by efficient, quantized models that run locally. Fully Offline AI All model inference happens on your device. Your chats are never uploaded, stored, or shared. Lightweight & Fast Models Llamatik supports multiple open-source models in GGUF format, including: • Gemma 3 270M • SmolVLM 256M / 500M • Phi-1.5 • Qwen 2.5 5B (quantized) • Llama 3.2 1B (quantized) Choose the model that fits your device and your needs. Built-In Model Downloader Download and manage AI models directly in the app: • Auto-download on first launch • Delete models you no longer need • Cancel ongoing downloads at any time Switch models at any moment to compare performance and quality. Clean and Modern Interface Llamatik features a simple, elegant chat interface built with Kotlin Multiplatform: • Smart suggestions • Model selector • Quick prompts • Smooth onboarding • Minimalist design Privacy First Your content stays entirely on your device. No login. No API keys. No cloud processing. The only data collected is anonymous analytics and crash reports to help improve app reliability. Powered by the Llamatik Library This app is a real-world demo of the Llamatik Kotlin Multiplatform library, supporting: • On-device LLM inference (via llama.cpp) • Embeddings • Vector store operations • Cross-platform support Developers can explore the capabilities through the app before integrating Llamatik into their own projects. Notes • Requires downloading at least one model to start chatting. • Internet is only used for model downloads and diagnostics. • All chat content is processed locally.

Decubed is the local-first AI workspace for the models, tools, memory, and agent workflows you choose. Most AI apps start with one model. Decubed starts with your environment. It gives you one place to connect model providers, work with local or remote model paths, keep durable threads, manage memories, and control which tools an agent can use. Use Decubed to: - Bring local, remote, linked-account, and custom models into one workspace. - Keep threads, folders, memory, tools, approvals, and run history attached to your environment. - Approve sensitive tool actions before they run. - Inspect tool calls, waiting states, and completed runs in the thread. - Configure agents around jobs with their own model, prompt, tools, skills, and memory policy. Decubed is designed for people who want model choice without giving up context or control. Run supported models on device when work should stay close to you, or connect hosted providers when a task needs remote capability. Bring the model. Keep the control.