Let me tell you what my AI stack looked like six months ago.
ChatGPT Plus. Claude Pro. Cursor Pro. GitHub Copilot. Perplexity Pro. OpenAI API credits for side projects. Grammarly. Notion AI. Slack paid tier. Zapier. Motion AI. Otter AI. TextExpander. Superhuman.
Fourteen tools. One thousand eight hundred and forty-seven dollars a year.And not a single one of them knew my name, my company, what I was working on, or what I did yesterday.
Every morning I opened Claude or ChatGPT and spent ten minutes re-explaining my startup. Cursor knew my code but not my Slack. Slack knew my team but not my calendar. Motion knew my calendar but not my priorities. Superhuman knew my email but not the Notion doc the email was about. Otter transcribed my meetings and dropped them into a void nothing else could read. Zapier was duct tape between all of it, and I was the one configuring the duct tape.
One afternoon I caught myself copying a Claude response into Grammarly to polish it, pasting it into Superhuman to send, then opening Motion to add a follow-up task. Four AI tools in a five-minute workflow, and I was still the smartest thing in the loop.
That's when I stopped.
***Why I'm writing this*
OpenHuman ships today. I'm the founder. This isn't an objective review and I won't pretend otherwise.
What I can do is tell you what pain this product was actually built to solve, how it works, and where it falls short. If you've read enough launch posts to know the pattern, that last part is the one that matters. Every product has a weak side. Founders who pretend otherwise are either new or lying.
***The four painpoints we set out to solve*
You are the integration layer.** Fourteen tools in my stack, zero talking to each other, and the glue holding it all together was my brain. My brain had better things to do. This is the modern knowledge worker's tax and somehow nobody is mad about it.
Your AI has no memory. Every model on earth is stateless. Claude, GPT, Gemini, all of them. The "memory" features that exist store a handful of bullet points. Any serious user blows past that in a day. You are permanently re-introducing yourself to the technology that's supposed to know you.
Your data is their data. Every AI tool you pay for sends your raw work to someone else's servers. OpenAI, Anthropic, Google, all of them. Your messages, your documents, your code, your screen activity. "We don't train on your data" is policy language. Policies change the second the board decides they should, and you find out in a blog post.
You're drowning in messages. Power users live in Slack, Telegram, Discord. 247 unread by noon. Most of it is noise. The handful of messages that actually matter are buried in the noise, and no amount of scrolling finds them in time.
OpenHuman was built to solve all four at once. On your machine, with your data, in one app. That's the whole pitch.
The walkthrough
Screen Intelligence captures your active window every few seconds and summarizes it locally. Over hours, you build a continuous record of what you actually did today. Per-app permissions, so you control what it sees. I can ask it "what was I working on before the standup?" and get a real answer instead of a guess. The first time I used this feature on myself I was embarrassed by how much of my workday was context-switching, and then I used the app to fix it.
Tradeoff: runs warm on older Intel Macs. Tuned for M1 and up.
Text Auto-Complete gives you inline completions across apps. Email, Slack, browsers, editors. The completions draw from your actual work context, not a generic writing model. I drafted a reply to a partner last week and the completion already knew the pricing we'd discussed with a different partner the day before. That is what AI was supposed to feel like the whole time.
Tradeoff: quality varies by app. Browser and native text fields are reliable. Some Electron apps with non-standard text handling are spottier.
Voice Intelligence handles dictation and voice chat with the assistant. For when you're pacing around thinking out loud, which is how most of my good decisions get made.
Channels connects to Telegram and Discord today, with iMessage and Slack on the near roadmap. The assistant reads, replies, searches, extracts action items, manages chats. On Telegram alone it supports around 70 operations covering the full Bot API surface.
This is what solved the 247-unread-messages problem for me. I stopped scrolling. I ask "what do I need to know from the last 12 hours?" and get a compressed answer. The decisions, the questions waiting on me, the contradictions, the threads that require my attention.
Tradeoff: iMessage and Slack are roadmap, not shipped. Apple integration is slow for the usual reasons.
Productivity integrations cover Notion, Google Drive, Gmail, and Calendar. Cross-source queries are the unlock. "What did we decide about launch timing?" returns a unified answer spanning Slack threads, Notion pages, and the email chain. Not three separate searches I combine manually in my head like it's 2019.
Tradeoff: initial setup takes 15-20 minutes to connect all sources. First-time indexing takes another 30 minutes depending on data volume.
The subconscious loop (this is the one I didn't see coming)
There's a background process that runs continuously while the app is open. It does recall loops across your indexed data, looking for patterns, contradictions, forgotten commitments, buried questions. 10,000+ thought loops a day for under $1 in inference cost.
Outputs look like: "You mentioned the design review to Sarah on Tuesday but haven't scheduled it." Or: "Your dev team agreed to ship Friday but your design dependency isn't ready until Monday, based on the Thursday standup."
I was deeply skeptical when one of our engineers proposed it. It sounded like the kind of feature that would be annoying in practice, pinging you constantly with things you didn't ask about. The threshold is tuned conservatively, so you get 3-5 of these per day, and about 80% of them are things I would have genuinely forgotten.
The first time this feature caught something I'd forgotten to follow up on with an investor, it paid for the whole app.
Rewards and referrals
There's a progression system built into the app. Streaks (7-day, 30-day). A Feature Maxi badge for using every major feature at least once. Power User tiers at 10M, 100M, and 1B cumulative tokens processed. Supporter roles for paid plans.
Connect Discord and the progression syncs. Exclusive channels, supporter badges, community access. Status and access, not cash for signups.
I was skeptical of progression in a productivity app. Usually feels bolted-on. In practice, the Feature Maxi badge was the thing that made me learn the whole app during my first week instead of opening two features and forgetting the rest existed. The streak counter has a mild nagging effect I don't hate. Your mileage will vary.
Referrals work the same way. You invite someone, they get a faster onboarding path, you both climb the progression. No money involved. Just access.
The weak spots, one more time
Linux support is live but rough. Daily fixes going out.
M1 Mac and up for smooth Screen Intelligence. Older Intel Macs work but warm.
iMessage and Slack integrations not live yet. Roadmap.
Local-first means no cloud backup by default. Export feature exists. Using it is on you.
Some Electron apps have spotty auto-complete quality. Working through it.
First-time setup is 30-45 minutes end to end. Worth it, not instant.
Code-editor use case is not our strongest suit today. Cursor and Copilot are still better for pure in-editor assistance. We may close that gap, but we're honest that it's open right now.
The ask
Download OpenHuman for Mac, Windows, or Linux: https://tinyhumans.ai/openhuman
If you want to follow the launch in real time, see what breaks, what ships, and what testers are saying, the launch thread is here: https://x.com/senamakel/status/2046266960707715277?s=46. Quote it, roast it, or just lurk. We're shipping 3-4 updates a day based on feedback, most of it coming from that thread.
Free trial on the paid tier for 2-3 months.
I'm terrified to ship this. I'm shipping it anyway.
Top comments (0)