Building in public · Open source · BSL 1.1

Your AI agents are
running unaudited.

Clampd is an open-source security proxy that intercepts, analyzes, and controls every tool call your AI agents make. Stop prompt injection, data exfiltration, and privilege escalation — before they happen.

Join waitlist View on GitHub
Coming Q2 2026 · Rust proxy + Python SDK
clampd proxy — what it looks like
# Agent requests tool call via LangChain
agent → sql.execute("DROP TABLE users; --")
 
✗ BLOCKED — destructive SQL detected
  rule: sql_injection_guard
  threat: prompt_injection → privilege_escalation
 
# Safe call passes through
agent → sql.execute("SELECT name FROM users WHERE id = 42")
 
✓ ALLOWED — read-only, scoped to user context
  latency: +1.2ms
The Problem

Agents have keys to everything.

AI agents call tools with full credentials. One prompt injection turns your helpful assistant into an attacker with insider access.

💉

Prompt Injection

Malicious instructions hidden in retrieved documents redirect your agent to execute harmful tool calls.

🔓

Privilege Escalation

Agents inherit broad OAuth scopes. One compromised chain grants access to delete databases, send emails, transfer funds.

📤

Data Exfiltration

Injected payloads silently exfil sensitive data through tool calls — API keys, PII, credentials sent to attacker endpoints.

🔗

Tool Chain Attacks

Multi-step agents amplify risk. A single poisoned step cascades through the entire chain with compounding damage.

👻

Shadow Agents

Developers spin up agents with API keys. No logging, no approval, no audit trail. Your CISO doesn't know they exist.

⏱️

No Kill Switch

When an agent goes rogue, there's no way to cut access in real-time. By the time you notice, the damage is done.

How It Works

One proxy. Total control.

Clampd sits between your agent and its tools. Every call passes through a 7-stage security pipeline — in under 2ms.

01

Intercept

Agent calls clampd proxy instead of the tool directly. Zero code changes — just swap the endpoint.

02

Identify

Clampd verifies the agent's identity via scoped tokens. No more shared API keys. Every agent gets least-privilege credentials.

03

Analyze

Intent analysis engine inspects the call payload. Pattern matching catches SQL injection, path traversal, data exfil attempts, and SSRF.

04

Enforce

Policy engine checks allow/deny rules against the agent's scope, the tool, and the payload. Violations are blocked instantly.

05

Deliver or Kill

Clean calls proxy through with minimal latency. Threats trigger the kill switch — revoke tokens, alert the team, log everything.

<2ms
Added Latency
7
Pipeline Stages
100%
Source Available
0
Code Changes
BSL 1.1 Licensed
Written in Rust
Sidecar & Helm
Sub-2ms Latency
Early Access

Clamp down before they break out.

Join the waitlist. Be the first to secure your AI agents when we launch.

No spam. Just launch updates. Unsubscribe anytime.