By Michael Saul. First published 2026-05-31, rewritten 2026-10-18 to match the product as it is today.
Reminder apps solve the wrong problem. Creating the reminder is easy. Following through is the hard part.
The reminder fires and I swipe it away. The habit tracker sits in an app I forgot to open. The todo list is where tasks go to sit for months. The calendar knows about the appointment but not about the form I needed to fill out before it. Every tool holds my intentions, and none of them keeps me moving.
So I built Ari around the place I already answer things: chat.
What Ari is
Ari is an AI accountability assistant you text. Ari manages your life, frees your time, and makes sure you reach your goal. You send a message on iMessage, Telegram, or Discord, or open the web chat at textari.chat. Ari is artificial intelligence, not a human.
You tell Ari what you are trying to do, what gets in the way, and what you need handled. Ari remembers enough context to make the next message useful, and follows up on the channel where you already reply.
What was hard
Follow-up that helps
A chatbot waits for you to open it. Accountability needs follow-up: scheduled reminders, habit check-ins, recurring pings, important-date nudges. A follow-up at the wrong moment is worse than silence, so check-ins have bounds and you can pause or change them any time.
Memory you can inspect
An assistant that remembers you becomes useful fast, and uncomfortable just as fast. Ari keeps layers: recent history, long-term memory retrieved by relevance, a structured profile, and short notes on voice and boundaries. You can ask what Ari knows, ask Ari to forget something, or export and delete your data.
Permissions
Tool access changes the risk. Reading a calendar differs from creating an event, and searching the web differs from sending an email. Integrations are opt-in and you can disconnect them later. Ari confirms provider actions through tool results before reporting that they happened.
Prompt injection
A webpage, email, or PDF can contain instructions aimed at the model. Asking the model to ignore them is a weak defence. The defence is structural, so each worker gets only the access its job needs:
- Web work sees only public information, never private memories or connected accounts.
- Integration work sees only the provider API for that job.
- Internal recall sees Ari’s own user-scoped data and never touches the internet.
- Group chats get a smaller tool surface and never receive private DM memory.
Scratch workspaces are wiped after each turn, and credentials stay outside generated code. The full design is on the Ari project page.
What Ari does now
- Tracks habits, logs done and missed, and keeps streaks visible
- Sends one-off and recurring reminders
- Remembers goals, people, routines, deadlines, and preferences
- Works with calendar, email, tasks, notes, and files after you connect them
- Searches the web, reads pages, and pulls YouTube transcripts
- Uses a live browser to check stock, prices, and hours and to fill out forms. You can watch and take over, and Ari asks before anything that spends money
- Builds small web apps and simple websites on request
- Handles files, screenshots, and voice notes
The same conversation works across iMessage, Telegram, Discord, and the web chat. The backend is a Python API with database-backed memory and background workers, and it runs on my homelab in Edmonton.
What is still rough
Ari is early. A follow-up can land at the wrong time, context can slip, tool calls can fail, and the model can state something wrong with confidence. Hard boundaries stay in place:
- Ari is AI, not a person.
- Ari is not therapy and not a crisis service.
- Ari does not make medical, legal, or financial decisions.
Try it and read more
- textari.chat: send the first message
- Pricing
- FAQ
- Guides and feature notes, including AI accountability partner and reminders that adapt when plans change
- Roadmap
- Source and other work: github.com/onionrings29
I wanted something that catches the thread when I drop it. That is Ari.