Persistent memory
Alice stores facts, preferences, and project context using memory skills and structured notes.
Alice Agent runs locally, remembers context, and handles code, tasks, and creativity without sending your workflow to the cloud.
Less context switching. More shipped work.
From install to first task in under a minute.
Run the one-line installer for macOS, Windows or Linux.
Point Alice at your model provider and optional skills.
Ask, code, automate, or create. Alice remembers and executes.
From small automations to complex creative projects.
Refactor code, write tests, review PRs, and manage infrastructure from a single chat interface.
Generate images, ASCII art, copy, diagrams, and polished HTML artifacts on demand.
Schedule cron jobs, reminders, and pipelines that run in the background while you focus.
Run tasks in isolated subagents so bad ideas never break your main environment.
From personal CLI to enterprise-ready multimodal workspace — here’s what’s coming.
From terminal to desktop, Alice runs tasks, keeps memory, and integrates with your favorite tools.
Alice stores facts, preferences, and project context using memory skills and structured notes.
Each capability is an Astro island: interactive only where it matters.
Alice stores facts, preferences, and project context using memory skills and structured notes.
Open sites, extract data, interact with forms, and verify results in real time.
Generate ASCII art, diagrams, music, copy, and code with a style of its own.
Create cron jobs, reminders, and pipelines that keep running while you rest.
Each subagent works in an isolated context to avoid conflicts and risks.
Selective hydration: animations and interactivity live only where they are needed.
Makers using Alice to ship faster.
I stopped bouncing between the AI chat, the terminal, and Notion. Alice keeps context across sessions, so I don't have to re-explain everything. The first week I handed off the repetitive deploy chores; I haven't touched them since.
I have an 8am cron that checks staging, summarizes open PRs, and drops it in Telegram before my coffee. Saves me half an hour of clicking every morning — exactly the half hour I never had.
80% of my job is writing queries and checking the result isn't lying. I describe the table I need and two iterations later the query works. I still review the SQL, but I don't start from scratch anymore.
I hand it my newsletter draft and it comes back in my voice, not AI voice. No 'moreover' every three lines, no obsession with lists of three. You can tell when something was written in a hurry — this wasn't.
I asked it to go through the old server's runbooks and rewrite them with what it found. Three hours later I had updated docs and no time left to read them. Memory keeps the decisions, not just the commands.
I hate doing code review on a Friday at 6pm. Now the agent pre-reviews diffs and leaves the serious comments on the board; I come in to decide, not to read everything. I still sign off on what I merge, but better informed.
Every Monday the team gets a summary of what got merged, what got stuck, and what's rotting in WIP limbo. I just check it doesn't say anything silly. So far it says fewer silly things than my own manual summary.
There was a regression suite nobody had touched since 2024. I asked Alice to bring it up, fix what was broken, and document why it broke. It ran in CI the next day. Nobody else touched anything.
Tickets arrive with technical context glued in from the latest conversation with the team. Not perfect, but a good deal better than 'make the button better' with nothing else.
The agent listens to the incident channel, builds the timeline, and posts it when we close. I used to write that timeline myself, at night, tired. Now I just review before publishing.
I ask it to research and come back with links I can verify. When it makes things up, it makes up the data, and that's how I catch it. In competitive reports, that's 90% of the work.
It wrote me a persistence module, tests and all. Swift isn't its strong suit, you can tell, but it sends a decent first draft and doesn't make me fight basic syntax. For prototyping it's gold.
Everything you need to know before installing.
Yes. Alice is open source under the MIT license and free to use, modify, and self-host.
The agent core runs on your machine. You choose which model provider to connect; your data never has to leave your system unless you configure external APIs.
A little. Alice is built for developers and technical makers who want a programmable assistant.
Absolutely. Skills are plain code and configuration files. Add integrations, cron jobs, memory sources, or custom tools.
Any model reachable via API: OpenAI, Anthropic, local Ollama / llama.cpp, OpenRouter, and more through the gateway.
Alice is local-first, memory-persistent, and designed to be extended by you, not locked behind a SaaS subscription.
Yes. Alice runs with Ollama, llama.cpp, vLLM, or any OpenAI-compatible server. No cloud, no API keys.
Only if you use a remote provider (OpenRouter, OpenAI, Anthropic...). With local models nothing leaves your machine. Sessions and memory live in ~/.alice/.
Alice is free and open source (MIT). You only pay for the model usage you choose — or nothing with local models.
Yes. Alice works as a library: import the agent and run prompts programmatically from your code.
Memory stores facts across sessions (who you are, your preferences). Skills are procedure documents the agent creates or installs to repeat workflows.
Each profile has its own memory, sessions, and skills. You can run several isolated agents on the same machine — even on the same repo with git worktrees.
Install Alice in one command and start delegating.
$ curl -fsSL https://alice-agent.stuko.dev/install.sh | bashOr download the desktop app
The desktop app bundles its own Python runtime — no separate install needed.See all releases →