Features

Everything below is free and works offline, except where a paid plan is named. Plans add online reach - they never take anything away.

Your AIs

Personalities you author

Create AIs from 18 personality archetypes with their own voices, appearances, and descriptions. Yours to shape, pause, archive, and restore - conversations included.

Response styles and modes

Conversational to thorough per AI; chat, report, and code modes per message, with automatic mode detection you can turn off.

Memory

A shared memory of you

Your AIs learn who you are from your chats - and Your Memory shows every fact with where it came from. Edit anything, forget anything, or pause learning entirely. Nothing is hidden from you.

Each AI remembers its own life

Every AI keeps its own memory of what you've done together and recalls the right moments when they matter - alongside memories you hand it directly.

Remember anything, anywhere

A Remember button under every reply, or highlight any text and keep just that. Choose the destination: one AI, all of them, or the project you're working in.

Project memory

Each project keeps a memory all your AIs share - conventions, decisions, context. Working AIs can save notes as they go (you approve each one), and sessions distill their takeaways when you finish.

Knowledge you give them

Add documents an AI should always know - it keeps its own copy and uses the relevant parts when they help.

Memory that survives

All of it - profile, per-AI, and project memory - is stored encrypted on your device, backs up to your Flowsta Vault, and comes back on a new machine.

Smart routing

The mode is the consent

"Auto - Offline Only" never touches the internet. "Auto - Online and Offline" may, by rules written in plain language in Settings. Pick a specific model and there's no routing at all.

The right model per question

Fit-aware picks on your device; hard reasoning can escalate to a stronger model; current-events questions can use web search with cited sources; health questions stay on your device by policy.

It always says why

Every reply names the model that answered, where it ran, and why it was chosen - with one-click second opinions: "Redo on your device" or "Try this answer online". Settings keeps a live list of recent routing decisions.

Your levers

Prefer fastest or strongest on-device. Privacy-first to freshness-first online. Per-category picks for which online model handles what, prices shown up front - and project work can stay entirely on your device.

Working with content

Files, images, and vision

Attach documents, spreadsheets, code, and images. Vision models read pictures; a one-time ~30 MB add-on reads scanned paper, fully on-device.

Verify sources

Answers about your documents can be checked claim-by-claim against the source, with the exact supporting quote - automatically or on demand.

Web search with citations

When you allow online use, current-events questions can search the web and cite their sources - with live progress while they research.

Doing work

Projects: agentic coding in chat

Open a project folder and your AI reads files, edits, and runs commands - always with your permission, every step recorded. A free add-on installs in one click.

Run in your terminal

Shell commands an AI writes can hand off to your own terminal, pre-filled at your prompt in the project folder - you stay the one who presses enter.

The inference engine

An OpenAI-compatible endpoint, serving your AIs

While the app is open, your machine serves a standard endpoint. Point agent frameworks like Hermes Agent and OpenClaw, coding editors, or your own scripts at it - and they get YOUR AIs, not a bare model. No API key needed on your own machine.

Connect your own server

Point the app at any OpenAI-compatible endpoint - a bigger model on your homelab box, a llama.cpp server on the LAN. It health-checks, measures its real speed, and its models join your picker.

The whole stack rides along

In either direction, everything applies: the personality you authored answers, memories inform it, routing treats connected hardware as a candidate, and every conversation lands in your signed records - external apps show an API badge on the Memory page.

Trust and your data

Signed conversation records

Every conversation is written into a tamper-evident record on your device - built on Holochain, the peer-to-peer foundation under Flowsta - and encrypted with a key only you hold.

Export and receipts

Export any conversation as a readable file with knowable contents - and optionally sign it with your Flowsta identity so anyone can verify it at flowsta.com.

Backup and device-loss recovery

Keys, conversations, AIs, and memories back up to your Flowsta Vault and come back on a new machine. Your data exports readable, keys included - no lock-in, by design.

Open source

AGPL-licensed, source on GitHub. Verify the privacy claims yourself - that's the point of making them verifiable.

Everyday comfort

Light and dark themes

The whole app in light or dark - switch anytime from the header menu, and every page follows instantly.

Help where you need it

Dismissible help tips explain each surface as you first meet it. Turn them all off in Settings - or bring back the ones you dismissed.

Models and performance

Model management without a terminal

Browse, download, and switch open models in the app, with hardware-fit guidance - what runs fully on your GPU, what will be slower, what won't fit.

Engines for your hardware

Vulkan and Metal out of the box, and a one-click CUDA engine for NVIDIA machines - the app picks safe defaults and steps down gracefully if your hardware protests.

Online frontier models (paid plans)

Frontier models from multiple providers with a monthly allowance and fair metered pricing. An add-on, never a dependency.

Wondering what you'd actually do with all this? See the uses →

Download free