Overview

A Hyperproxy is a credential-injecting forward proxy that injects real API credentials into HTTP/S requests without exposing them to the clients making those requests.

You give your Hyperproxy your secrets once. Your agents, scripts, and tools connect through it using a simple per-proxy username and password. Your Hyperproxy swaps in the real credentials before forwarding the request upstream.

What a Hyperproxy solves

Modern AI agents and automated scripts need to call third-party APIs — GitHub, OpenAI, Stripe, Slack — but distributing real API keys to every process creates a sprawling credential management problem. Rotating a single key means touching every deployment. A compromised key means immediate exposure.

Your Hyperproxy centralises credentials in an encrypted store. Your agents see only proxy credentials — short-lived, revocable, scoped per-proxy. If an agent is compromised, you revoke its proxy credential. The underlying API keys are unaffected.

Two Operating Modes

Every Hyperproxy runs in one of two modes, chosen at creation. Pick the one that matches how you work — you can always create a second proxy in the other mode.

Key Hyperproxy features

• Credential injection — Authorization: Bearer {$GITHUB_PAT} becomes the real token at the proxy layer
• CEL-based firewall — per-request allow/deny/hold rules with full access to method, URL, host, path, headers, and body sample
• Human-in-the-loop (HITL) — hold suspicious requests and surface them in the dashboard for manual approval
• Per-proxy credentials — each proxy has its own username/password pairs; compromise one, revoke one
• IP allowlist — restrict which source IPs can connect to a proxy
• Access logs — every proxied request is logged with action, rule matched, and bytes
• Metrics — built-in metrics endpoint on each proxy for monitoring and observability
• Multi-tenant — each account gets isolated proxies; secrets never cross tenant boundaries
• Hyperproxy DNS resolver (Pro+) — a private DNS resolver that blocks malware, trackers, and custom domains at the DNS layer using curated blocklists and Shannon-entropy DGA detection
• Credential vaults — isolate credentials per upstream service with host-bound ICAP injection and predefined service templates
• Secure sharing — issue expiring, use-limited, TLS-terminated proxy capabilities without revealing API keys
• Cross-platform CLI — manage proxies, vaults, and shares from macOS, Linux, and Windows with OAuth 2.0 device flow authentication
• Expiry notifications — receive email and/or SMS alerts at configurable lead times (up to two weeks in advance) before any secret or proxy credential expires

How a Request Flows

# Your agent (no real credentials needed)
curl -x "https://<proxy-host>:443" \
     -U alice:proxy-password \
     -H "Authorization: Bearer {\$OPENAI_API_KEY}" \
     https://api.openai.com/v1/chat/completions \
     -d '{"model":"gpt-4o","messages":[...]}'

#  ↓  Hyperproxy resolves {\$OPENAI_API_KEY} from encrypted store
#  ↓  Evaluates firewall rules (if firewall mode)
#  ↓  Forwards with real Authorization: Bearer sk-proj-...
#  ↓  Returns response to your agent

Architecture

Each Hyperproxy runs on its own dedicated virtual machine with a dedicated IP address. No resources are shared with other tenants. Each proxy consists of:

• Enterprise-grade proxy engine — handles HTTP/HTTPS and TLS interception with battle-tested reliability
• Hyperproxy credential sidecar — manages secret injection, per-credential authentication, and CEL rule evaluation
• Encrypted secret store — AES-256-GCM encrypted database; secrets never leave the machine

Each proxy supports up to 26,872 unique authentication credentials simultaneously on the same dedicated IP — one per port in Hyperproxy's multi-port range. The control plane manages proxy lifecycle and pushes configuration changes to proxies through a secured internal API.