Intelligent Internet's Genii Texts You First as an Open-Source iMessage AI

Most AI assistants sit inside a browser tab, waiting for you to remember they exist. Every session starts from zero, you carry the context, and you have to go find them. Genii, from Intelligent Internet, works the other way around. It lives as a contact in your iMessage, remembers everything you've told it, and texts you first when something needs attention.

A contact, not an app

Setup is deliberately frictionless. No app to download, no account to create. Enter your phone number, wait a few minutes, and a new contact appears in iMessage. From that point, you interact with it exactly like texting a friend.

The hosted version is live at genii.ii.inc, with iMessage first and more surfaces on the way. Genii runs inside iMessage through the Linq app, a service that lets software send and receive iMessages programmatically. Pricing hasn't been announced, but the full codebase is open source under Apache 2.0, so you can self-host it for the cost of your own API keys.

Six things it can actually do

Genii is built around six core capabilities:

• Initiative: Genii watches the streams you've connected and acts on what it finds: the flight open for check-in, the pull request waiting on review, the promise you made in passing three days ago. When something needs you, it texts you first.
• Persistent memory: Every conversation becomes memory: names, preferences, decisions, the way you like things done. During idle hours it runs a background consolidation process the team calls "dreaming," connecting the note from March to the plan you made last night and pruning what stopped mattering.
• Integrations: Link Gmail, Google Calendar, Slack, GitHub, Google Maps, and Google Drive. Each connection is both a tool and a source of signal: Genii can search your inbox, draft replies, summarize the Slack channel you've been ignoring, and surface the document you only half-remember.
• Voice mode: Text "Talk mode" and the conversation shifts to live voice, same assistant, same memory, in real time.
• Its own computer: Every Genii instance gets a real machine and a real browser, isolated in the cloud. It shops with your signed-in accounts, writes documents, runs research, and ships code while you're away.
• Group chat: Add Genii to a group conversation and it participates like anyone else: settling debates, booking the restaurant, turning "we should do a trip" into an actual itinerary.

The architecture underneath

Genii runs a Python 3.11+ backend with a Node.js 22+ frontend, using either SQLite for local development or Postgres for production persistence. The agent runtime executes inside an E2B sandbox, a cloud-isolated virtual machine with a real desktop browser, so when Genii browses the web or runs code on your behalf, it does so in a contained environment that never touches your local machine directly.

The repository provisions OpenClaw 2026.6.5 in that sandbox and accepts your own OpenAI API key when you create the agent. Credentials stay encrypted in the local backend; the agent runs in a sandbox you control.

For long-running tasks that don't fit a chat thread, Genii delegates to II-Agent, Intelligent Internet's open-source agent built for extended autonomous work across research, code, data analysis, and content generation. Genii stays in the thread; II-Agent does the digging. When the work finishes, the result arrives as a message in the same conversation where you asked.

Integrations are handled through Composio, a third-party service that manages OAuth flows and webhook triggers for Gmail, Google Calendar, Slack, GitHub, Google Drive, and Google Maps. Write actions, sending an email, creating a calendar event, posting to Slack, stop at a consent gate and wait for your approval before executing. The approval is recorded, and enforcement is built into the architecture rather than left to model behavior.

Self-hosting: what you need

To run it yourself, the quick-start path looks like this:

make setup
# Build the Genii desktop template into your E2B account:
SANDBOX_E2B_API_KEY=... uv run python e2b_templates/genii/build_template.py --name genii-desktop
# Add keys to .env, then:
make doctor
make dev-basic

The template build is a one-time, paid E2B operation. After that, open http://localhost:1430, use local developer sign-in, and create your Genii with your OpenAI API key. Three run modes are available depending on how much infrastructure you want:

• Browser mode (make dev-basic): SQLite only, no Docker required. The fastest way to get started.
• Postgres-backed (make infra && make db-migrate && make dev): adds a local Postgres instance for production-like persistence.
• Full callback mode (make tunnel && make dev): exposes a public HTTPS URL for voice handoffs, Composio OAuth triggers, scheduler delivery, and Linq webhooks for iMessage.

One caveat from the README: this first milestone targets local development and does not yet promise production self-hosting. Telegram, Discord, billing, managed model credentials, and operator tooling are on the roadmap but absent from this release.

Presence over capability

The broader AI assistant market has spent years competing on model quality. Genii's bet is that the bottleneck was never intelligence; it was presence. ChatGPT, Claude, and Gemini all reset context between sessions and require you to initiate. Genii is built around the opposite assumption: an assistant that knows you on Tuesday because it knew you on Monday, and speaks first when something needs you.

The memory architecture is the key differentiator. Rather than a simple conversation log, the background consolidation process connects disparate pieces of information across sessions and prunes stale context, closer to how a good chief of staff operates than how a chatbot works.

Intelligent Internet is positioning Genii as the consumer layer on top of their broader open-source agent stack, with the GitHub repo already structured around a Community Edition and Enterprise Edition split, signaling a commercial tier is coming. The hosted product is the easiest entry point today. Teams that want to customize the memory and integration layer can pull the full open-source stack and bring their own E2B account and OpenAI key.