Wirken is an open-source gateway that sits between your chat apps and your AI agents, and it starts from the right assumption: any agent can be talked into working against its owner. It connects Signal, WhatsApp, Telegram, Matrix, Discord, Teams, Google Chat, iMessage and a web chat, each as its own process holding only its own credentials.
- Credentials live in an encrypted vault and get added at the edge, so the agent and model never see them.
- Shell commands run in a no-network throwaway container. No sandbox, no execution.
- Reads run freely. Changes ask once and remember. Deletes, installs and credential access ask every time.
- Every action goes into a signed, hash-chained log before it runs.
Help Net is upfront about the gaps: MCP servers and the LLM client still make their own outbound connections, injection detection flags but doesn’t block, and in a shared channel one person’s approval covers everyone until it expires.
My take
This is the design I want for my own bots: least privilege, ask before anything destructive, and an audit trail you can verify. Read the limits before you trust it, and export the audit log before you ever wipe the data folder.
God Bless, and May the Source Be with You!
