# Agent Auth Gateway
https://agent-auth-gateway.onrender.com

Agent Auth Gateway is the authorization layer for always-on AI agents: budget caps, action allowlists and denylists, human approval queues, and kill switches, enforced through a drop-in proxy endpoint. Your agent sends every tool/API call to POST /v1/proxy with its agent id, action name, and estimated cost; the gateway returns allow, deny, or require_approval, and logs every evaluation.

Key features:
- Agent registry with per-agent policy: daily and monthly spend caps, action allowlist/denylist, risk tier (low/medium/high).
- Drop-in proxy: POST /v1/proxy with {agent_id, action, estimated_cost}; returns allow / deny / require_approval. All evaluations logged.
- Human approval queue: pending items approved or denied from the dashboard or via a signed email link (no login needed).
- Kill switch: one click disables an agent; the proxy immediately denies its calls.
- Consent-gated policy rules (Pro): one-click ARTICLE5_CONSENT_GATE template gates image-generation tool calls (generate_image, *image*) on consent attestation metadata (context.consent_attested=true + context.attested_by) — missing attestation is denied or routed to the approval queue, optionally tripping the kill switch; deterministic, never judges image content. Tied to the EU AI Act Article 5 enforcement wave, 2 December 2026.
- Dashboards: per-agent spend vs caps, action counts, approval history, anomaly alerts (spend spikes, new action types, caps nearing).
- Notifications: webhook per workspace (Pro) plus an in-app inbox; approval emails when SMTP is configured.
- Pro instruments (server-side 402 gated): Burn Forecast & Auto-Throttle (per-agent days-until-cap-exhaustion with graceful degraded policy mode), What-If Policy Simulator (replay 200 real evaluations against an edited policy), cross-agent burn leaderboard. Team: Anomaly Step-Up Lockdown (agents auto-lock on anomaly; every action needs a human until released).

Pricing: Free ($0 forever: 1 agent, basic caps, no approval queue). Pro $49/month (10 agents, approval queue, webhooks, alerts, burn forecast, policy simulator). Team $149/month (unlimited agents, 5 seats, audit export, full API, anomaly lockdown).
Target user: teams running always-on AI agents who need spend control and human oversight on every action.
Guides: /guides hosts in-depth articles on AI agent cost control and authorization policies (spend caps, approval queues, allowlists vs denylists, burn forecasting, anomaly step-up lockdowns).
Comparison: /vs/agentcost is an honest AgentCost alternative comparison (verified 2026-10-12): where AgentCost wins (forecasting depth, MIT self-hosted community, integrations) vs where Agent Auth Gateway wins (self-serve $49/mo governance, action-level policies, kill switch, no-login approvals, no self-hosting).
Policy playground: /playground runs the real fail-closed policy engine in the browser (no signup): paste an action log or pick a preset (runaway pay loop, rogue delete) and get live allow / deny / require_approval verdicts with reasons.
How to buy: message us on WhatsApp at https://wa.me/5016691053 or email avilanatanael640@gmail.com to get a license key, then redeem it on the /billing page. Card payments coming soon.