The governed loop, in four steps.
Agents connect, read, and propose. You approve. Here's the trust gate, end to end — try it below.
One path, every agent, every time.
Click through each step, or let it play. The gate stays closed until a human opens it.
Connect over MCP
Any MCP-capable agent connects to your workspace with a scoped token. One protocol, no bespoke integrations to maintain.
Watch the loop run.
Connect, read, propose, approve — the same gate, in your terminal. Click a step to replay it.
Guardrails that aren't optional.
Proposals, not writes
Agents submit reviewable diffs. You see exactly what would change, in context, before anything happens.
Scoped, least-privilege access
Each agent gets a token scoped to specific boxes and capabilities — read here, propose there, nothing more.
Auditable and reversible
Every step lands on an append-only log, with full version history and one-click rollback on every object.
From any agent to your source of truth.
Agents connect over MCP and propose changes. Nothing is written to your governed store until you approve — and every step is versioned and audited.
Your agents
Claude
ChatGPT
Cursor
Copilot
Windsurf
Gemini
Cline
VS Code+ any MCP client
Poggle — the trust gate
Your governed store
Neon Postgres
RedisVersion history · append-only audit
Ready to put a human in the loop?
Connect your first agent in minutes. No credit card needed.