The archive · AI & Models · Product decision · 2025–2026
OpenHuman's readable-memory bet: 10K GitHub stars in a weekend, 34K by end of Q2 2026
TinyHumans' local-first desktop agent OpenHuman: 118+ OAuth integrations, human-readable Markdown memory, 10K stars in a weekend after May 2026 launch.
TinyHumans (Tiny Humans, Inc.) · OpenHuman
What the business is
OpenHuman is an open-source (GPL-3.0) local-first desktop AI agent: it connects 118+ third-party tools (Gmail, Notion, GitHub, Slack, Calendar, Drive, Linear, Jira) via one-click OAuth, stores a user's long-term memory as a human-readable Markdown vault on-device, routes to any model including local Ollama, and ships a desktop mascot that can join calls.
Starting capital：Not disclosed; Runa Capital's ROSS Index lists no funding raised as of end of Q2 2026 (founded 2025, San Francisco).
How it started
Founder Steven Enamakel has said the project came from watching his father struggle for three hours to set up an open-source AI agent — API keys, YAML files, a never-opened terminal — until both gave up. TinyHumans, Inc. was incorporated in San Francisco in November 2025; first code commit to the repo was 2026-01-27. The team built with Tauri + Rust core + React instead of Electron, for lower memory use and native Windows support.
What happened
OpenHuman launched publicly in May 2026 and went viral: 10,000 GitHub stars in one weekend (compared with 62 days for OpenClaw), repeated top rankings on Product Hunt, and widespread coverage in Chinese tech media (36kr, Tencent Cloud) questioning whether it is 'an agent or an OS'. By 2026-06-10 the repo had 30,969 stars, 134 contributors, and ~1,269 commits in May alone; Runa Capital listed it as #10 among open-source startups by Q2-2026 star growth at 33.9K stars. The default memory pipeline, TokenJuice, claims ~80% token-cost savings, and the project's README positions it as 'not AGI, but an architectural step closer.'
How it ended up
Still scaling: as of 2026-06-30 OpenHuman was at ~34K GitHub stars and releasing frequently (v0.61.x by early July), with no funding round or exit reported; the company had not disclosed revenue or a commercial product beyond the free open-core desktop app.
Background
OpenHuman is an open-source, local-first desktop AI agent from San Francisco startup TinyHumans. It connects 118+ SaaS tools through one-click OAuth, keeps a long-term Memory Tree in human-readable Markdown on the device, routes requests across cloud and local models, and gives the agent a desktop mascot that can join meetings. Founder Steven Enamakel says the idea came from watching his father spend three hours failing to set up an open-source agent with API keys and YAML files — OpenHuman was built so that no one needs to do that.
The project hit GitHub in May 2026 and went viral immediately: 10,000 stars in a single weekend (OpenClaw had needed 62 days for its first 10K), repeated top placements on Product Hunt, and 30,969 stars with 134 contributors by June 10, 2026. Runa Capital's ROSS Index ranked TinyHumans #10 among open-source startups by Q2 2026 star growth, at 33.9K stars. Coverage in Chinese tech media framed it as the third act after OpenClaw and Hermes — an agent that behaves more like a personal operating system than a developer tool.
The company's bet is that the agent war will be won at the desktop entry point, not in model quality: a UI-first app with managed OAuth, a built-in cheaper agent-tuned model, Windows support from day one, and a memory layer users can read and edit. As of end of June 2026 OpenHuman had not disclosed any funding round, revenue, or exit; it was still early-beta, shipping rapid releases, with GPL-3.0 open-core as its distribution strategy.
What has to be true
- The wedge was accessibility: everyone else (OpenClaw, Hermes) assumed technical users; OpenHuman targeted the 99.99% who can't edit YAML, with native installers and one-click OAuth.
- The moat bet is the memory layer: an on-device, human-readable Markdown vault that compresses a user's whole digital life into context on first sync — hard for cloud-first rivals to replicate.
- The timing was deliberate: Tauri 2 and falling token costs made a Rust desktop agent with always-on memory economically viable in early 2026.
- Distribution was open-core and free: GPL-3.0 code plus a free default model created the virality (10K stars in a weekend) that no paid launch could have bought.
What can be applied
Once a category is proven, the winning move can be removing every setup barrier — installers, one-click OAuth, cheaper routing, readable memory — and letting accessibility go viral.
Aftermath
As of 2026-09-02, OpenHuman is live and actively developed (releases through v0.61.x, with recent versions adding faster memory flows and concurrency improvements). Star growth has continued past the Q2 spike, though no later company milestone — funding, revenue, or partnership — had been announced. The project remains early-beta, with known OAuth regressions and Windows packaging issues flagged by community analysts, and its main strategic risk is competition from closed desktop agents (e.g. Anthropic's Cowork) if their memory layer is opened.
Sources
- OpenHuman刷屏硅谷，开源、免费、能接100多个应用，但我越用越后怕
- 4 个月 31K stars：OpenHuman 用可读记忆杀入桌面代理
- Top trending open-source startups in Q2 2026 (ROSS Index)
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card