Clampd ships as a suite. App devs wrap a client in one line. AI-coding-tool users drop in a hook. Security teams run the dashboard; ops lives in the CLI. All of them share the same gateway and the same 287 detection rules underneath.
If you want a turnkey on-ramp instead of picking individual products: sign in to app.clampd.dev, install the Python or TypeScript SDK with one line, get a free agent slot, and you're guarded. No infra to run, no Redis or Postgres to configure. Free under 25 agents and 500k tool calls/month. Upgrade only when you need OIDC SSO, compliance reports, or higher volume.
There are five integration paths into the Clampd gateway, plus operational, testing, and reporting surfaces. Pick by where the AI lives in your stack:
| Static analysis (CI, pre-merge) | clampd-discover |
| Code you control | SDK (Python, TypeScript) |
| IDE coding agents | clampd-guard (Claude Code, Cursor) |
| AI coding agents in CI/CD | clampd-action (Claude Code) |
| MCP servers (any transport) | MCP proxy |
| Network edge (no code change) | clampd-wire |
| Operations and audit | Dashboard, CLI, reports, replay |
| Security regression | clampd-redteam |
SDK and clampd-wire are alternatives, not substitutes. Most enterprises end up running both: SDK for code they own, wire for shadow AI and unwrapped traffic. Discover ties them together by mapping every AI surface in the codebase.
Firewall for AI coding agents that run inside your pipeline. When Claude Code runs as a CI step, every tool call it makes (Bash, Read, Write, WebFetch) is routed through the gateway and verified before it executes. One secret, no infra to run, built on clampd-guard.
Clampd Discover: shadow AI inventory. A command-line scanner that maps your AI surface in seconds. clampd scan <path> walks any repo, monorepo, or directory and finds every place an LLM SDK, framework, or agent runtime is used: OpenAI, Anthropic, Gemini, Bedrock, Vertex, Mistral, Cohere, LangChain, LangGraph, LlamaIndex, AutoGen, CrewAI, Semantic Kernel, Ollama, Transformers, vLLM, llama.cpp, and MCP servers (Cursor, Claude Code, Continue, Aider). For each finding, Discover names the model, identifies whether the call is already wrapped with Clampd or running as shadow AI, and recommends a Clampd bundle that would govern it. Output is JSON, SARIF (GitHub Code Scanning), or human-readable.
Wire Mode: catch every LLM call at the network edge, no SDK. Intercept every LLM call leaving your environment at the network edge. No SDK wrap, no code changes. Set HTTPS_PROXY once, install the per-org Clampd CA cert, and every prompt, tool call, and response from OpenAI, Anthropic, Gemini, Bedrock, Vertex, or self-hosted models passes through your governance pipeline. Same rules, same Cedar policies, same bundles as the SDK path. Streaming responses re-encoded in flight without buffering. This is how you cover shadow AI: agents Discover finds get routed through the same enforcement as the ones you wrapped by hand.
Maintained compliance evidence: HIPAA, GDPR, SOC 2, PCI-DSS templates kept current as the regulations evolve. Scheduled monthly/quarterly delivery. Activity inventory: agents observed, tool descriptors observed, MCP endpoints with traffic, policy violations grouped by agent/time/category. The raw audit data lives in your own ClickHouse on every tier and is queryable directly. clampd-reports is the maintained alternative for teams who don't want to keep their templates current as compliance frameworks evolve.
Pick any historical agent decision and reconstruct the full call: tool descriptors, params, matched rules, session-pattern flags, scope token issued, LLM-as-Judge verdict (if fired), final policy decision, downstream response. When an alert fires (kill, repeated denials, behavioural anomaly), Clampd auto-assembles every related decision, agent activity, scope grant, and delegation hop into a single timestamped, signed incident packet ready for legal hand-off, insurance claims, or regulator review. Multi-cluster federation across distributed deployments. Long-term retention beyond dashboard defaults. The underlying ClickHouse audit data is yours on every tier; clampd-replay is the templated workflow plus chain-of-custody signing infrastructure on top.
The gateway behind every product runs in your environment of choice. Same code, same feature parity. Pick the operational fit:
See the setup page for the deployment-mode-specific install steps.
We publish a real subset of the code and ship the detection core as a licensed binary. Every tier runs the full security pipeline, we never gate whether an attack gets blocked. Here's exactly what's open, what's source-available, and what's proprietary.
OSI-approved, no usage restrictions. Wrap your agent and read every line.
Full source + tests published. Your security team can review every line before deploying.
Ships as licensed binaries. Free on the Design Partner tier; the license unlocks scale and governance.
The license gates scale and governance, agent count, SSO/RBAC, multi-tenant, support, never detection. Browse the published code at github.com/clampd/clampd.