Vigil sits between your autonomous agent and the real world — blocking dangerous actions, logging everything, and escalating to humans when it matters.
No spam. Just early access when we launch.
Your agent never touches Stripe or your database directly. Every action goes through Vigil first.
No AI guessing. Pure logic: amount limits, rate limits, SQL injection blocks, PII filters. Always predictable.
Every action logged to ClickHouse — what the agent tried, which rule fired, and the outcome. Full forensic replay.
Agent wants to charge $600? Vigil pauses, pings Slack or Telegram, waits for your Approve / Reject before continuing.
Track token usage and cost per agent, per action. Know exactly how much each autonomous decision costs.
One POST request from your agent. Works with any framework — LangChain, AutoGen, custom — regardless of LLM provider.
Your data stays in your infrastructure. Deploy with docker-compose in minutes. No vendor lock-in.
Before executing any real-world action, your agent calls Vigil's check endpoint.
# Before: agent calls Stripe directly — no safety net # After: agent asks Vigil first POST https://vigilgate.tech/api/v1/check X-Api-Key: your-api-key { "agent_id": "billing-agent-v1", "target": "stripe.charge", "action": "create", "payload": { "amount": 150.00, "currency": "USD" } } // Rule: amount_limit $100 → BLOCKED ← 403 { "status": "blocked", "reason": "amount 150.00 exceeds limit 100.00 USD" } // Rule: require_approval $500 → ESCALATED ← 202 { "status": "escalated", "hitl_id": "550e8400-..." } // No rules triggered → PASSED ← 200 { "status": "passed" }
Whether you're a developer shipping agents or a person using AI assistants daily.
You're building autonomous agents for clients or internal use and need a safety layer before go-live.
You use AI assistants (n8n, Zapier AI, ChatGPT Actions) and want guardrails without writing code.
Start free. Scale when your agents do.