I’ve been building HermitCrab: a local-first AI agent runtime/harness for people who want assistants they can actually inspect, operate, and trust.
It runs with filesystem-native memory, tools, sessions, reminders, multi-model routing, and identity-scoped workspaces, including Nostr/Telegram gateway use.
Beta4 is out. I also made a short explainer video.
Repo: https://github.com/talvasconcelos/hermitcrab
If you care about local-first agents, personal AI infra, or safer agent operations, I’d love feedback.
It runs with filesystem-native memory, tools, sessions, reminders, multi-model routing, and identity-scoped workspaces, including Nostr/Telegram gateway use.
Beta4 is out. I also made a short explainer video.
Repo: https://github.com/talvasconcelos/hermitcrab
If you care about local-first agents, personal AI infra, or safer agent operations, I’d love feedback.
2
