Integration architecture
One authorization layer for every tool, framework, and LLM provider
Agent frameworks, protocol surfaces, and CI pipelines all call the same gate. Wrap execution once, receive PERMIT, DENY, or SILENCE with a signed receipt.
The gate pattern · 3 lines
const gate = new TrigGuardGate({ apiKey: process.env.TG_KEY });
const receipt = await gate.authorize("db.delete", {
recordId: "user_01",
});
if (receipt.decision === "PERMIT") {
await database.delete("user_01");
}System map
Every stack converges on the same authorization API
Framework choice does not change the trust path. Tool routers, MCP servers, and function calling all evaluate through TrigGuard before side effects.
LangChain
CrewAI
OpenAI Tools
MCP
MCP GitHub Actions
TrigGuard Gate
POST /execute
PERMIT
DENY
SILENCE
Ecosystem
Works with the stacks you already run
Recognizable integrations, not admin dashboard tiles. Same authorization surface regardless of runtime.
Supported surfaces
Frameworks and protocols, one enforcement model
Agent frameworks
- LangChain Live
- CrewAI Live
- PydanticAI Live
- AutoGPT Beta
Protocols & transport
- MCP Live
- OpenAI function calling Live
- gRPC / Protobuf Live
- HTTP / JSON Live
Deploy the authorization layer
Reference patterns, adapter specs, and runtime docs for wiring TrigGuard into your stack.