Greyline is a reverse proxy that detects autonomous AI agents probing your API and deploys a counter-agent to interrogate, delay, or mislead them, before a single request reaches your real infrastructure.
Free tier available. No credit card required.
How Greyline works
Greyline detected a LangChain agent, scored it 94/100, and routed it to the Bouncer. What follows is the actual conversation. Your real API never received a single request.
Your API subdomain points to Greyline via CNAME. Every inbound request is inspected at the edge. Zero latency for legitimate traffic.
Framework UA match, missing browser headers, HTTP protocol anomalies, superhuman timing, and 15 more detectors run simultaneously. Score computed in <1ms.
A Claude-powered counter-agent takes over the session. The agent believes it's talking to your API. Your real origin receives nothing.
The Bouncer interrogates, stalls, or poisons. It burns the agent's context window, budget, and clock time. Up to 50 turns.
Simulated demo session: representative output from the Bouncer on a detected LangChain agent:
Setup
Point your API subdomain at Greyline's edge. That's the entire setup. We provision TLS, handle routing, and start detecting immediately.
Add one CNAME record. TLS provisions automatically.
20 signal detectors run on every inbound request across three false-positive tiers.
Detected agents are routed to a counter-agent. Your origin server is never touched.
Confirmed agent signatures propagate to all Greyline deployments automatically.
Greyline fires alerts by email or webhook when something real shows up.
Greyline only stores data from confirmed agent traffic. Legitimate requests are proxied and immediately discarded — no logging, no storage, no exposure of your real infrastructure or request bodies. This is architecturally enforced: only the bouncer path writes to the database.
Defense modes
Greyline adapts to your threat model. Start with Interrogate to waste agent time. Graduate to Poison when you want agents to walk away with confidently wrong data.
The Bouncer engages the agent in a bureaucratic conversation — every turn burns context window and API budget.
Responses are artificially delayed 8–30 seconds, causing upstream failures for agents on tight timeouts.
Return convincing HTTP 200 responses — the agent believes it succeeded while Greyline logs the full session.
Greyline generates schema-valid but entirely fabricated API responses — agents walk away with confidently wrong data.
Included on every plan
POST to Slack, PagerDuty, or your SIEM the moment an agent is detected. Every payload is HMAC-SHA256 signed. Pipe detections into your incident response workflow without polling the dashboard.
See exactly what the agent did: fetched your schema, probed auth endpoints, enumerated user records. Every session shows the full URL sequence with automatic step classification — spec discovery, auth probe, data access, enumeration.
Allowlist your CI/CD pipelines, uptime monitors, and internal tooling by IP, CIDR, or User-Agent prefix. Those requests bypass scoring entirely — zero false positives from your own infrastructure. Changes propagate in under 60 seconds.
Every detection has a public URL — agent score, framework attribution, signals that fired, attack path timeline, and full interrogation transcript. Share with your team via one link, no login required. Attach to Jira tickets, Slack threads, or security reports.
Every Monday morning: total sessions, confirmed agents, top detected framework, week-over-week trend, and a link to the most suspicious session of the week. Async awareness without opening the dashboard.
Confirmed agent fingerprints are anonymized and shared across all Greyline deployments. When one customer catches a new agent pattern, everyone benefits. Business tier customers can query the full feed via API — integrate threat intelligence directly into your SIEM or security tooling.
Who uses Greyline
SaaS APIs
Data APIs
Financial APIs
E-commerce APIs
Internal APIs
Any HTTP service
Pricing
Every plan includes the full detection engine. Upgrade when you need the counter-agent features that make interception useful.
Free
Forever free. No credit card.
Pro
Everything in Free, plus:
Business
Everything in Pro, plus:
Bouncer turns are Claude API calls made during agent interrogation. When your monthly allocation is reached, detected agents receive a silent honeypot response. Your API stays protected; interrogation pauses until next month.
Trust signal
Add a Greyline badge to your docs or status page. It's a live SVG that updates with your real blocked-agent count. Some operators find it deters low-effort agents entirely.
Your badge preview
Common questions
Configure a webhook endpoint in Settings → Webhook. Greyline sends a signed JSON payload (HMAC-SHA256 via X-Greyline-Signature) the moment a session exceeds your score threshold. The payload includes the session ID, score, framework attribution, and a shareable URL. Works with Slack, PagerDuty, or any HTTP endpoint. See the webhook docs →
No. Add your runner IPs, CIDR ranges, or User-Agent prefixes to the Trusted Automation allowlist in Settings. Matching requests bypass scoring entirely — not logged, not intercepted. Changes propagate in under 60 seconds. See the allowlist docs →
Yes. Every session has a permanent public URL at /s/:sessionId showing the full interrogation transcript, signals that fired, and attack path timeline. No login required. Use the share button in the session drawer, or grab the shareUrl from any webhook payload. Good for Jira tickets, Slack threads, and security incident reports. See the sharing docs →
Yes. Every Monday at 9am UTC, Greyline emails a digest covering the past week on your hostname: total sessions, confirmed agents, unique IPs, average agent score, top detected framework, and week-over-week trend. Available on all tiers. No configuration needed.
Yes. The session summary email fires after each confirmed agent session (score 61 or above). It includes the threat score, framework attribution, action taken, interrogation turn count, attack path (first 5 URLs), and a shareable link to the full transcript. Enable it under Account → Email notifications in the dashboard, or click Manage email preferences in any Greyline email. See the email docs →
Yes. Most emails are optional. Go to Account → Email notifications in the dashboard to toggle individual types on or off — weekly digest, session summary, first-agent alert, CNAME reminder, and 7-day silence alert can each be disabled independently. Every optional email also includes a one-click unsubscribe link (no login required). Cap warnings cannot be disabled since they signal your protection has gone inactive. See the email docs →
Runs on Cloudflare Workers. Your traffic never touches a foreign server. Privacy & security policy →