Ostler vs OpenHuman.

OpenHuman is an open-source desktop app from tinyhumans.ai, built in Rust on Tauri, shipped under GPL-3.0, and led by Steven Enamakel out of Dubai. It runs on macOS, Windows and Linux. It is a thoughtful piece of engineering and a worthy competitor. Ostler is a single-machine personal AI Hub for macOS, with an iOS app, sold as a one-time $99 Hub purchase plus a $9.99 per month Ostler Pro subscription (first 30 days of Pro included). The architectural difference is the load-bearing point of this page.

The “118 integrations” question

OpenHuman’s public marketing says “118+ third-party integrations”. That is true in one sense and worth unpacking in another.

118+ services are callable by the OpenHuman agent. They are brokered through Composio, a third-party integration platform that publishes a catalogue of 1,000+ OAuth-connected services. OpenHuman curates a subset of those for tool-call use.

However, only 3 of those 118+ services auto-ingest data into the local Memory Tree: Gmail, Notion and Slack. This is visible in the source: init_default_providers() in src/openhuman/composio/providers/registry.rs registers exactly those three native providers with their own ingest pipelines and entity extractors. The remaining 115+ are passive: the agent can call them on demand, but they do not flow into your local knowledge store unless you write the bridge yourself.

Both numbers are true depending on what you call an integration. If “integration” means “the agent can call this service for you”, 118+ is right. If it means “this service’s data lands in your local memory so the agent can reason over your history with it”, the number is 3.

Ostler picked the second meaning. We auto-ingest from the surfaces a Mac user actually lives in (iMessage, WhatsApp, Mail, Calendar, Contacts, Photos, Safari) plus 20 GDPR-export pipelines for the platforms whose data you can legally pull. The number is smaller; the depth is meaningfully different.

Where OpenHuman wins

Cross-platform. If you run Windows or Linux, OpenHuman runs there. Ostler is macOS-only on the Hub, with the iOS app on iPhone. We picked Apple-native depth over cross-platform reach.

Open source. OpenHuman is GPL-3.0. You can read every line, audit the privacy claims for yourself, and fork it if you disagree with where the project is going. Ostler is closed source. That is a real trade.

Bigger community. 9,246 GitHub stars in under three months and a 30-strong contributor halo is the kind of momentum that compounds.

Tool-call breadth on demand. If you want your assistant to be able to reach 100+ work tools through one OAuth broker, Composio is the right answer for that. Ostler does not aim to replicate this.

First-mover on the consumer brand. A Product Hunt #1 launch is a real moment. They earned it.

Where Ostler wins

Pure-local default. Every meaningful Ostler query runs on your Mac. Ollama hosts the model, Qdrant stores the vectors, Oxigraph stores the graph triples, the wiki renders locally with MkDocs. Pull the ethernet cable and the system still works. Cloud routing is purposefully absent. Personal data never leaves the Mac.

Apple-native lived data. Ostler reads iMessage chat.db, Mail via Full Disk Access, Photos, Contacts, Calendar, and Safari history. These surfaces exist only on a Mac, and they are where the most personal and useful relationship data sits. A cross-platform Tauri shell cannot reach them at the same depth from Windows or Linux.

One-time Hub payment. $99 once for the Hub, with Ostler Pro at $9.99 per month (first 30 days included). The Hub side of the value never has a subscription. Your data, the local intelligence, and Ostler chatting locally over your snapshot keep working whether or not you continue Pro.

Zero per-query cost. Local inference means no per-query API bill. You pay for electricity, not tokens. Heavy users see this on the margin; users with sensitive data see it on the trust margin.

iOS app. Native iOS app at launch: wiki browser, life timeline, people search, suggestions, quick capture. OpenHuman is desktop only.

The architectural argument

OpenHuman and Ostler are not making the same bet in different colours. They are making different bets, for different customers, on the same general thesis (“personal AI with persistent memory is the right shape for the next ten years”).

OpenHuman’s shape fits an always-on, work-tool-heavy customer: someone who wants their agent to live across Notion, Linear, Slack, Gmail and 100+ other SaaS tools, with a meeting-agent that can join Google Meet calls, and an LLM router that picks the cheapest sufficient cloud model for each task. The hosted backend at api.tinyhumans.ai is load-bearing; it makes the cross-tool, cross-provider, cross-platform story possible. The Memory Tree on disk is a useful cache, but the meaningful compute happens upstream.

Ostler’s shape fits someone whose most valuable data is intimate rather than corporate: years of iMessage threads, family photos, the email archive that traces every relationship of the last twenty years, the calendar that tells the truth about where time actually went. For that customer, the architectural commitment matters more than the integration count. The single-machine model is the privacy guarantee. If the data never has to leave the Mac to be useful, it never does.

Both products are honest about their trade. OpenHuman’s privacy and security docs are clear that LLM calls, web search and TTS are proxied through their backend. Ostler’s commitment goes the other way: every service runs locally on the customer’s Mac, full stop.

The world does revolve around you.^™