Run AI agents on hardware you already own.

Core complete, pre-customer. Stripe billing integrated. Free, Pro, and Enterprise tiers live. First paying customers targeted within 6 months.

Apple Silicon + chat platforms + VM-level isolation — the only stack that does all three.

• Agent deployment across 5 chat platforms — WhatsApp, Discord, Slack, Telegram, Mattermost. Core complete.
• MicroVM isolation per agent — ~125ms boot, <10MB overhead. Apple VZ on macOS, Firecracker on Linux. Running.
• Security enclaves — RBAC, DLP/tokenization, audit logging, secret management. Running.
• 100+ MCP tools — messaging, scheduling, VM control, fleet management. Production.
• Air-gapped mode — vsock-only communication, no internet egress. For deployments that cannot touch the public network.

skiff builds on the rest of the tmc labs stack. cove provides the microVM isolation substrate. mlx-go runs inference locally on Apple Silicon when the deployment calls for it. mcp is the tool-access layer every agent speaks to reach its environment. langchaingo is the Go LLM framework a skiff agent is typically written against — stewarded by tmc labs, community-owned, MIT licensed.