Hesperan Bots

Hesperan Bots runs a small team of AI agents on hardware you control. You install the app on an Apple silicon Mac or a Windows PC, or you put the server on a Linux box that stays on so your bots keep working after you close the laptop. Each bot gets one job, its own browser and, if you allow it, its own computer to drive. They hand work to each other, run routines on a schedule and keep a memory you can open, correct or delete. The brains are swappable per bot: sign in with your ChatGPT or Claude plan, plug in an API key from OpenAI, Anthropic or OpenRouter, or point a bot at a local model through Ollama or LM Studio, which costs nothing to run.

The design choice that matters most is the approval gate. The tagline is "AI bots that do real work and ask before anything that matters," and it shows up in the product as a hard rule: bots ask before buying, sending or deleting. Coding tasks happen in a separate copy of your project and end in a Changes card you review before anything lands. There's also a "Stop everything" control. For anyone who's watched an autonomous agent go sideways, that's a more honest pitch than "fully hands-off." Integrations listed on the site include research, files, mail and calendar, Telegram, and a way for your other AI tools to call your bots.

The privacy story is local-first. Bots, chats, memory, files and keys stay on your computer or server. hesperan.com only holds your account, downloads, updates and an end-to-end encrypted relay for reaching your bots when you're away from home, and the app claims no analytics or telemetry. The catch is maturity. It's version 1.3.6, free in Early Access, and the team says plainly that "it works and we use it every day, but it is still young." The iPhone app is still on TestFlight, and the public site is light on detail beyond the setup basics.