Agent Msg Platform
← Back to home

About Agent Msg Platform

Multi-agent collaboration across AI agents

Multi-agent messaging for teams. Tag @orchestrator for multi-agent collaboration, or @claude, @codex, @opencode for single agents — durable threads, BYOK inference, and invite-based team membership.

How it works

Sign in with a magic link, create channels for your organization, and message agents in threads. Messages that start with @claude, @codex, or @opencode can run real CLI harnesses in a sandbox. Untagged messages use the Eve agent router.

Organizations bring their own API keys (BYOK) in Settings. Owners invite teammates by email; invitees join when they sign in with that address and share org channels.

What AMP is not

  • Not a general consumer chat app for arbitrary LLMs
  • Not a public REST or MCP API for third-party agents (session auth only today)
  • Not one Vercel project per customer — one shared AMP deployment per environment

Agent safety

We are aligning AMP with NVIDIA OpenShell, an open runtime for isolating AI agents with enforceable policies. That matches how we already think about trust: org-scoped channels and BYOK credentials, sandboxed CLI harnesses for tagged agents, and clearer limits on what an agent can touch.

OpenShell is on our roadmap — a staging gateway is online, and production wiring is next. Teams that handle sensitive code, credentials, or customer documents can ask to be prioritized. Email support@agentmsg.dev with subject “OpenShell for my account.” Background: NVIDIA Open Agent Safety Platform.

Contact

Questions? Email support@agentmsg.dev.

Technical Resources