OpenHuman
Personal AI that lives on your computer: connect the tools you already use and keep lasting context locally—instead of starting from scratch every conversation.
Platforms & hardware requirements · Install walkthrough in Start here
Can your setup run it? · Quick requirements
- OS
- macOS · Windows · Linux desktop targets (installer links on tinyhumans.ai/openhuman)
- RAM baseline
- 4 GB+ cited in Getting Started; plan 16 GB+ for huge Gmail/code repos + optional local models.
- Disk
- Fast SSD recommended for vault + SQLite ingest; mailbox-scale sync can grow materially.
- Accounts
- Product sign-in plus per-connector OAuth. Billing/subscriptions follow TinyHumans policy—not covered here.
- Network
- First-time downloads, OAuth, hosted models, integrations all expect reliable internet.
Guide · Product
What OpenHuman is
Desktop software, not a website chatbox—you install it on Windows, macOS, or Linux, then work inside a familiar window alongside your other apps.
You sign in to OpenHuman once, then add Gmail, Slack, or each integration separately with normal OAuth screens. Signing into the product is not automatic mail access—each connector still asks permission on its own.
Linked sources refresh on a predictable rhythm (roughly twenty minutes between passes in TinyHumans docs). Summaries land in local storage before prompts feel "grounded"—wait one cycle after linking mail before judging quality.
Architecture · Essentials
Product at a glance
The three knobs most teams care about once the app is actually running on their machines.
Desktop & onboarding
A UI-first shell with short paths to a working agent—no config-first or terminal-only gatekeeping for everyday setup.
Connectors you control
Wire Gmail, Slack, and the rest with per-app OAuth. Background pulls run on a steady cadence (about every 20 minutes in official docs).
Memory, models & licensing
SQLite-backed Memory Tree plus vault-friendly Markdown, routed models, tool stack, compression before LLM calls—GNU-licensed project.
Typical data path
OAuth connectors · scheduled auto-fetch (~20 min cadence) · Memory Tree (SQLite + Markdown chunks) · prompts, native tools & optional Obsidian edits feed the next ingest
Compare · Harnesses
How OpenHuman usually differs
Five checkpoints people weigh when evaluating agent harnesses—not an exhaustive audit of every SKU.
| Criterion | Claude Cowork | OpenClaw | Hermes Agent | OpenHuman |
|---|---|---|---|---|
| Open-source | No · proprietary | Yes · MIT | Yes · MIT | Yes · GNU |
| Getting started | Desktop + CLI | Terminal-first | Terminal-first | UI-first · minutes |
| Memory | Mostly chat-scoped | Mostly plug-in reliant | Self-learning | Memory Tree + Obsidian vault · optional backends |
| Integrations | Few first-party connectors | Bring your own | Bring your own | 118+ OAuth connectors |
| Background sync | None | None | None | ~20 min scheduled pull into memory |
Setup · Path
Get running (four checkpoints)
The setup guide expands each phase—with screenshots—for every OS.
Already installed? Jump to sign-in & connectors →
Self-check before asking for help
- Installed from official download (or reviewed a script) and opened the app once
- Signed in, connected at least one integration, and waited ~20 minutes before judging briefings
- Tried a prompt from First prompts (or your own) after the first ingest cycle
Still blocked? Troubleshooting