OpenBot
Self-hosted AI coworkers, each with its own computer and a policy gateway
Emerged finds new products worth a look, daily — what’s new today
It drives real desktops, browsers, iPhones and Android phones from one pip install with your own model key, but its shell tool runs commands without asking first.
AutoCua runs a team of agents that look at the screen, click, type, browse and write code. It covers macOS, Windows and Linux desktops, five browsers through its own extension, Android emulators and phones, and iPhones or iPads from a Mac host. Install is pip plus a setup script per system, with a key for one of ten model providers. MIT, built and maintained by Ashish Yadav. The README is frank about limits: no automated tests, no sandbox or approval step for shell commands, and runs of one task can take different routes.
More like this
DashClaw
Approval gates and an audit ledger for coding agents left running unattended
CTRLRun
An open-source safety layer that checks every AI agent action before it runs
AgentRun
Open-source DSL that freezes repeated agent work into typed, inspectable steps
Legion
AI agents that act inside your Elixir app by writing sandboxed Lua
ColliePWA
Answer your coding agents' prompts from your phone, over your own tailnet