// features
Not a chatbot.
An assistant with an architecture.
Every feature below exists for one reason: to give you a personal AI that's actually yours — memory, brain, voice, and all.
01 · MEMORY
Memory you can open, read, and edit.
grAIm doesn't have a vague "personalisation layer." It has a real memory system — structured entries with IDs, tags, priorities, and token counts, stored in a database on your phone.
- ◆ Browse everything. The Memory Browser shows every fact, skill, and reference your assistant holds — searchable, filterable, editable, deletable.
- ◆ 40+ skills out of the box. The app comes bundled with over forty ready-to-go skills, each with its own customisable preferences.
- ◆ Priorities & budgets. Memories are marked Critical or Optional with token costs, so your assistant loads the right context without blowing out your API bill. A smart selection system automatically pulls in the memories that match what you're asking, so responses stay tailored without spending your whole budget.
- ◆ Granular by design. Memories are stored and indexed individually — not buried in wikis or markdown files — so context is built fresh for every prompt, on demand. No compaction, no "context engineering". Assistants built on RAG and other outdated approaches hit scaling limits that cap both what they remember and who they can be. That's the AI in grAIm: granular AI memory.
- ◆ It compounds. Mention your project once, and weeks later your assistant connects it to a new conversation — automatically.
02 · YOUR LLM, YOUR CHOICE
One assistant. A whole pipeline of brains.
Bring your own keys for Claude, ChatGPT, Gemini, DeepSeek, Z AI, OpenRouter (200+ models) — any OpenAI-compatible API. But grAIm goes further than "pick a model":
💬 Response model
The brain that talks to you. Use a frontier model for quality or a budget model for everyday chat.
🧩 Extraction model
A separate model distils topics and learnings after each response — so memory-building can run on a cheaper model.
⚖️ Judgement model
A different-provider model reviews memory-pipeline decisions — catching correlated mistakes a single model can't see in itself.
🖼️ Image generation model
Generates images from your prompts — for avatars, mockups, or anything else you ask it to draw.
👁️ Image analysis model
Vision-capable — analyse photos, screenshots, or documents you share in the conversation, and translate text from images in any language.
🔊 Sound analysis model
Coming soon — audio and environment analysis, so your assistant understands more than just words.
Switch any of them anytime. Your memory travels with you — zero lock-in.
03 · VOICE-FIRST
Talk to it like a person.
Talk to your assistant anywhere — in the car, on a walk, cooking dinner. Speech-to-text and text-to-speech can run on-device, so your hands stay free and the conversation flows naturally. No typing — just talk.
- ◆ Hands-free flow. The assistant listens, thinks, and answers aloud — with a live animated state so you always know what it's doing.
- ◆ Pick the voice. Choose from multiple on-device voices and TTS providers — Gradium, ElevenLabs, OpenAI and Google — to personalise your assistant's sound. More providers are added over time; send us your request.
- ◆ Private by physics. Keep audio processing on-device and there is no server to send your voice to.
- ◆ Live translation. Speak to your assistant in a foreign language and it translates into yours — built in, ready to go.
Button colours
🔵 Blue — idle, ready
🔴 Red — listening
🟠 Orange — thinking
🟢 Green — responding
04 · WIDGETS
Glanceable. On your home screen.
Your assistant holds a lot of information. Widgets put the important parts right on your home screen — no app to open, no question to ask. Think of them as keyhole previews into specific memories.
📅 Calendar
Today's date and your upcoming reminders, pulled live from your assistant's memory. Tap to access calendar view.
📈 Ticker
Live prices for anything you want to track — crypto, stocks, commodities. Just tell your assistant what to track, from what source, in what currency, and it sets it up. No menus, no fixed lists — fully customisable to whatever matters to you. Tap to refresh.
🧠 Memory
A keyhole preview of any memory your assistant holds — a skill, a saved reference, a list. List-type memories display as an interactive checklist you can tick off from your home screen.
05 · MAKE IT YOURS
Name it. Style it. Give it a face.
Your assistant isn't a product mascot — it's yours. Shape everything about how it looks, sounds, and behaves.
- ◆ Animated .gra avatars. State-aware animation — idle, listening, thinking, speaking — designed to be efficient enough that it won't cook your battery. Prefer something simpler? Plain image avatars keep things simple too.
- ◆ Personality shaping. Direct, warm, dry-witted, formal — you define how your assistant carries itself.
- ◆ Its own name. Ellie, Jarvis, Max — the name you choose is the identity it keeps, across every conversation.
06 · SKILLS & TOOLS
Teach it once. It knows forever.
Skills are reusable procedures your assistant learns — by you talking to it, not by you writing code. Correct it once on how you like something done, and that becomes a rule it follows.
- ◆ Built-in tools. Web search, calendar, lists, reminders, research — the assistant routes requests to the right tool automatically using our smart selection system.
- ◆ Skills are memories. Stored in the same inspectable memory system, tagged Critical when they matter, and retrieved by the same smart selection system based on what you're doing — you can read exactly what your assistant learned.
- ◆ grAIm Marketplace coming. Share and discover skills, identities, avatars, and other grAIm-compatible assets — free or paid — on the roadmap.
07 · PRIVATE BY DESIGN
You decide what it sees.
Toggle by toggle.
Granular permissions control exactly what device context your assistant can access — date & time, locale, battery, network — each rated by sensitivity, each off until you say otherwise.
See the full privacy architecture →How grAIm compares
The short version: everyone else keeps your memory. We never touch it.
Comparison reflects typical products in each category as of 2026.