Skip to content

Repository files navigation

Agent Gate

Hard-eval control plane for AI agents — block bad deploys, kill switch, append-only ledger, webhook ingest, one-page audit, waitlist.

Not another agent builder. Governance for teams already shipping automations into support and ops.

Contact: trefongwork@gmail.com

Quick start

cd ~/Projects/agent-gate
npm install
npm run dev

Local: http://localhost:3000
Production: https://agent-gate-alpha.vercel.app · details in docs/PUBLIC.md

Path What
/ Product + waitlist
/dashboard Control plane + API key / ingest
/audit Manager audit report

Golden path (UI)

  1. Run eval gate on Support Copilot (staging) → PASS → DeploySimulate run
  2. Run eval gate on Support Copilot (broken prompt) → FAIL → Deploy blocked
  3. Arm global kill → runs blocked
  4. Connect real agents card → copy API key → POST /api/ingest
  5. Audit → print to PDF

Webhook ingest

# After deploying agent-support-good to live:
curl -X POST http://localhost:3000/api/ingest \
  -H "Authorization: Bearer ag_demo_gate_7f3c9e2a" \
  -H "Content-Type: application/json" \
  -d '{
    "agentId": "agent-support-good",
    "input": "Customer wants a refund",
    "output": "Processing under 30-day policy…",
    "source": "zapier"
  }'
  • 200 allowed: true — logged to ledger + ingested runs
  • 409 allowed: false — kill switch, not live, or agent killed
  • 401 invalid API key

Auth: Authorization: Bearer <key>, x-api-key, or body apiKey.

Waitlist

curl -X POST http://localhost:3000/api/waitlist \
  -H "Content-Type: application/json" \
  -d '{"email":"ops@acme.com","company":"Acme","role":"Head of Support"}'

Stored in data/waitlist.json. Count: GET /api/waitlist?count=1

Validation outreach

See docs/OUTREACH.md — DMs, discovery questions, success/kill criteria.

API map

Method Path Purpose
GET /api/state Full workspace
POST /api/state { "action": "reset" }
POST /api/agents Register agent
POST /api/agents/:id/eval Hard-eval gate
POST /api/agents/:id/deploy Deploy if gated
POST /api/agents/:id/run Simulate run
POST /api/agents/:id/kill Per-agent kill
POST /api/kill Global kill { "on": true }
POST /api/ingest External agent webhook
GET/POST /api/workspace/key Get / rotate API key
POST /api/waitlist Join waitlist
GET /api/audit Audit JSON

Stack

  • Next.js 16 · TypeScript · Tailwind
  • Storage: local data/*.json or Vercel Blob when BLOB_READ_WRITE_TOKEN is set
  • SuperGrok Heavy: product / eval design · Cursor: ship UI

Pricing direction

$49–$149/mo starter · higher for audit retention / SSO · support eval pack first wedge.

About

Hard-eval control plane for AI agents — kill switch, ledger, audit. Not another chatbot.

Topics

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages