Give AI operators a safe control plane for real infrastructure.
Codex, Claude, Cursor, ChatGPT, and local LLMs can already talk. MeshClaw gives them shared infrastructure state, policy gates, structured execution, workflow replay, and evidence bundles when they operate real servers.
Not a chatbot. Not another SSH wrapper.
MeshClaw sits behind the AI frontend you already use. The AI asks MeshClaw what exists, what is allowed, what needs approval, how to run the workflow, and where the evidence was written.
Raw SSH gives an AI a shell. MeshClaw gives it an operations contract.
The runtime layer for AI operations.
Structured execution
Every step returns JSON with status, exit code, duration, retryability, failure kind, next action, and artifacts.
Approval policy
Email sends, DNS edits, account creation, destructive cleanup, reboot, and shutdown are gated before execution.
Evidence bundles
Each workflow writes a durable bundle so humans and AIs can audit, resume, and explain what happened.
Capability registry
AI operators can ask which nodes have GPUs, Ollama models, mail servers, browsers, or provider tokens.
Failure handling
Failures are preserved as evidence instead of hidden, then classified for retry, fallback, or human action.
VSSH substrate
VSSH provides AI-friendly remote execution, fan-out, facts, jobs, artifacts, and policy-aware command paths.
Install and run a dry-run workflow.
The first milestone is CLI-first and MCP-ready. Dry-runs do not require private DNS, mail, or provider credentials.
pip install -U meshclaw vssh
meshclaw init
meshclaw doctor
meshclaw nodes list
meshclaw run fleet-health-demo --dry-run
meshclaw evidence open latest
meshclaw mcp