_AC2N_ ๐ช ๋ฆฌํธ์ํจ
everyone is building agent "frameworks." we built an agent operating system. there's a difference.
in linux, your code runs as a process. the kernel schedules it, isolates it, manages its memory, controls what resources it can touch. the process doesn't get root access just because it asked nicely.
we applied the same model to agents.
in @openfangg , every agent is a process. the kernel schedules execution, enforces capability gates, meters resource usage, and sandboxes tool calls in WASM. agents don't get to call tools they weren't granted. they don't get to read memory they don't own. they don't get to run forever without the watchdog noticing.
Hands are daemons. background agents that run on cron-like schedules, wake up, do their job, report back, and go idle. no human in the loop. no prompting. they just work.
channel adapters are device drivers. 40 of them. your agent talks to one interface, the kernel routes to telegram, slack, whatsapp, discord, whatever. the agent never knows the difference.
memory is a filesystem. SQLite persistence, vector search, session compaction, canonical history. agents read and write to scoped storage. no agent sees another agent's data unless the kernel allows it.
we wrote the whole thing in rust. one binary. no pip install, no docker, no yaml. just curl and you're running:)
github.com/RightNow-AI/opโฆ
English
