How to make Instinct 10x more useful

I’ve been into personal assistants since mid 2025. Lobu started as one, basically OpenClaw but multi-tenant. I’ve run Muse, OpenClaw, Hermes, bunch of others people argue about on X.

The models are smart enough already since the start of this year but somehow most people have hard time finding good use-cases for personal assistants. The assistants just couldn’t reach my stuff: what’s going on in my life, and the tools I actually use. My mail, my calendar, my machines, a browser that’s already logged in to the right places. Without that you have a chatbot that’s polite about not being able to help.

Instinct is the one I’ve had the least trouble with. No setup, no config files, no babysitting a server. I signed up and it worked, which wasn’t true of any of the others. It also doesn’t wait for me to type. The heartbeat, a periodic check-in, keeps an eye on what I’m doing and pings me when something needs me, where the others only answered when I remembered to ask.

how instinct works

Instinct runs its own model. Each task gets an ephemeral sandbox, and it has a vault for secrets, so a credential never sits in a chat. It has few direct integrations of its own. The browser use is good enough that it still gets a huge amount of work done.

That’s also how it does QA. It drives a real browser like a person would, clicks through the flow and checks the result, so when it says something works it has seen it work. I have a list of user flows that I run it with Instinct occasionally and it created 20 pull requests to fix the issues it found so far.

Why do I use Instinct with Lobu?

I don’t hand Instinct my accounts. In my setup it doesn’t connect to anything directly. Mail, calendar, devices, the browser, all of it goes through Lobu connectors, so there’s one place where I grant access and one place where I take it back. That way my ChatGPT, Claude, Instinct, all the agents see the same connectors and tooling.
Instinct can’t connect to WhatsApp, and it can’t drive my local computer. Without Lobu it would miss a big chunk of what’s going on in my life. Through Lobu my machines can open pages in my own Chrome session and do whatever Instinct wants done there. I also hooked Codex and Claude up to the same computers, and Lobu keeps the shared context across all of them, so they don’t step on each other.
It builds a full timeline of my events: mail, calendar, what I do in the browser, what runs on my machines. That’s how the assistant can see what I’m up to without doing tens of tool calls. I use automations to create goals like task triage and Instinct sees all these events and proactively reaches out to help or remind me with the work.

A personal assistant shouldn’t have to call tens of services to build a task tracker, a separate system should work outside of the agent loop so the data is available to your agent whenever you need. That’s why we built Lobu as a context layer that can be integrated to any agent. You can run it on Lobu’s cloud or self-host it on your own machine.

It also means I’m not locked in. I experiment with OpenAI’s agent and Muse too, and since all the context sits in Lobu I can switch between assistants whenever I want. I can even get them working together on the same thing, because none of the context belongs to any one of them.

Instinct alone is a smart thing that can only talk. Lobu alone is plumbing nobody is using. Together it’s an operator.

How Lobu connects Instinct to my accounts, tools and machines

on the road

On a trip to Turkey, Instinct looked up travel options and made a few purchases while my coding sessions kept running on the Mac mini back home. When Claude needed me, the heartbeat pinged me and I answered from my phone.

Looking for rentals is where Lobu clicked for me. Sahibinden, a large classifieds site in Turkey, requires a local login from abroad, so Instinct couldn’t get in. Through Lobu it used my own browser session and opened the listings. As far as the site could tell, it was me browsing.

what I found poking at it

Instinct is a closed product, so I did what I do with anything I depend on. I probed it. It resisted when I asked it to describe its own architecture, but a few things came out, and I posted them as field notes on X.

burak emre

@bu7emba

apparently @noahrshinn's instinct uses kimi k2.6. it resisted pretty hard when i tried to get it to reveal its architecture, but here are a few things i found: - uses ephemeral sandboxes rather than persistent ones, unlike grok bot and muse - relies heavily on residential proxies - it can dynamically create secret vault entries for you to populate through the web app. they claim ai doesn't see secret but it technically can. - token limits are daily and suspiciously high right now, at least 100m+ - their rl-tuned model seems heavily optimized for browser use - so far, it’s the only multiplayer ai i’ve seen that lets your bot talk directly to your friends’ bots through a trusted-person network it's a good product but is it 10x better than @bot or @Muse to beat their distribution? i'm afraid not.

21Sep 13, 2026View on X ↗

Try it with your own assistant

You can ask your assistant to connect itself to Lobu. Paste this to it:

Prompt

Connect to Lobu for me. Read https://lobu.ai/auth.md and follow it, using my email address (ask me for it if you don't have it) so it's my own account and shared context. Lobu will email me a magic link. I'll click it and approve. Then save the token in your memory or secure store so you can reuse it without asking me again.

The magic link is the one manual step: you click it once and approve, and your assistant does the rest. The refresh token lasts 30 days. It works with any assistant that can make web requests.

If you’d rather set it up yourself, create a workspace, connect the sources you want your assistant to reach, then use the assistant connection guides. You can also do it from the terminal with the Lobu CLI. Start with one source and one task, like checking what’s waiting in your inbox.