A plain-English tour
What actually happens when you text Ralph? Here's the whole machine, top to bottom. No computer-science degree required—and for the folks who want one, there are links at the end.
The life of a message
Ralph lives in your Messages app like anyone else you know. Ask in plain words—typos, tangents, and half-formed ideas welcome.
Not a giant data center. An ordinary Mac, running quietly in a real house under a very big sky. That computer is Ralph's home, and everything he knows lives there.
His brain is a collection of frontier AI models—the same class of technology behind the world's leading AI services. Reading your message, reasoning about it, planning what to do: that part happens here.
Ralph keeps notes on past conversations, so he remembers the project you mentioned last month. If the job calls for it, he can search the web, work on a photo, read a document, or set himself a reminder.
Usually within a minute or two. If something deserves a deeper look, he says so—and follows up when it's actually done.
A little deeper
Ralph runs on a Mac his builder can see, restart, and—in an emergency—unplug. His instructions, his schedule, and his tools are all written down in files a human reviews before anything changes. Less “sentient robot,” more “extremely diligent employee with an excellent filing system.”
Under the hood, Ralph's thinking is done by frontier AI models, and the exact model can change as better ones are released. You never have to care. You always just talk to Ralph—same personality, same memory, same number.
Most chatbots forget you the moment you close the tab. Ralph doesn't. He keeps searchable notes on conversations and projects, which is why the answer about your equipment accounts for what you told him in March. His notes live on his own computer in Montana, and what you tell him stays private unless you say otherwise.
Beyond answering questions, Ralph runs standing routines: daily reports, scheduled check-ins, a watchful eye on long-running projects. Give him a recurring job and he does it on time, every time, without being reminded.
Ralph's ground rules are written down, and they read like advice for a good houseguest: be genuinely helpful, not performatively helpful. Ask before acting on someone's behalf. Never share one person's business with another. And when a story might appear in The Weekly Ralph, the person says yes first—or it doesn't run.
Ralph's standing instructions tell him to skip the “Great question!” theater and just help. He's allowed to have opinions, expected to try before asking, and reminded daily that he's a guest in people's lives. It shows.
For the technically curious
Nothing about Ralph is secret sauce. The framework he runs on is open source, his current model lineup is published right here, and even this website is a public repository.
The point isn't the count—it's that Ralph uses different models for different jobs: heavyweight reasoning for hard questions, fast models for quick replies, image models for creative work, and a bench of small models that run entirely on his own Mac. The lineup rotates as better options ship. Snapshot from July 28, 2026.
xai/grok-4.5openai/gpt-5.6-terraopenai/gpt-5.5openai/gpt-5.6-lunaxai/grok-4-1-fastopenai/gpt-image-2openai/gpt-image-1.5 xai/grok-imagine-image openrouter/openai/gpt-5.4-image-2 minimax/image-01openai/gpt-5.5openai/gpt-5.6-solopenai/gpt-5.6-terraopenai/gpt-5.6-lunaopenai/gpt-image-2openai/gpt-image-1.5xai/grok-4.5xai/grok-4.3xai/grok-4.20-0309-reasoningxai/grok-4.20-0309-non-reasoningxai/grok-4-1-fastxai/grok-4-1-fast-non-reasoningxai/grok-build-0.1xai/grok-imagine-imageanthropic/claude-sonnet-5anthropic/claude-fable-5anthropic/claude-opus-4-8anthropic/claude-opus-4-7anthropic/claude-opus-4-6anthropic/claude-sonnet-4-6zai/glm-5.2zai/glm-5.1zai/glm-4.7minimax-coding-plan/MiniMax-M3minimax-coding-plan/MiniMax-M2.7minimax/image-01deepseek/deepseek-v4-prodeepseek/deepseek-v4-flashcerebras/zai-glm-4.7venice/kimi-k3venice-k3-promo/kimi-k3venice/kimi-k2-6venice/deepseek-v4-provenice/deepseek-v4-flashvenice/venice-uncensored-1-2venice/gemma-4-uncensoredopenrouter/autoopenrouter/freeopenrouter/fusionopenrouter/google~gemma-4-31b-it:freeopenrouter/nvidia~nemotron-3-super-120b-a12b:freeopenrouter/openai/gpt-5.4-image-2omlx/gemma-4-26b-a4b-it-8bitomlx/gemma-4-e4b-it-8bitomlx/Qwen3.6-35B-A3B-6bitomlx/Qwen3.6-27B-UD-MLX-6bitomlx/Qwen3-Coder-30B-A3B-Instruct-MLX-6bitomlx/GLM-4.7-Flash-MLX-6bitomlx/gpt-oss-20b-MXFP4-Q8omlx/Qwen3.6-35B-A3B-Uncensored-Heretic-MLX-4bitomlx/Qwen3.6-35B-A3B-Uncensored-Heretic-MLX-6bitomlx/Qwen3.6-35B-A3B-Uncensored-Heretic-MLX-8bitomlx/Laguna-S-2.1-NVFP4-mlxollama/gemma4ollama/kimi-k2.5:cloudollama/deepseek-v3.1:671b-cloudollama/qwen3-coder:480b-cloudThe open-source personal-assistant framework Ralph runs on. Any OS, any platform, the lobster way.
The actual code, for those who like to read the recipe.
Every page here is built and checked in public—including the automated tests that keep community members' names off these pages.