Skip to main content

TrustGate is an AI agent gateway: a control plane between agents and the models, tools, and MCP servers they call. Point an OpenAI, Anthropic, or Responses client at it. Then route, authenticate, apply policy, and record every call — from the NeuralTrust console or the control-plane API.

How it works

1

Authenticate the caller

A consumer presents an API key, or a JWT that TrustGate validates against the user’s IdP. The application never holds the provider credential.
2

Apply policy and route

TrustGate evaluates rate limits, budgets, tool allow-lists, and guardrails, then the consumer’s registries handle the upstream, with fallback and smart routing when you configure them.
3

Record the hop

Tokens, cost, latency, and routing attempts go to traces and OpenTelemetry.

Create a gateway

Register a provider, create a consumer, and send the first request.

Install a data plane

Run TrustGate in your environment when prompts cannot leave your network.
Before-and-after diagram showing separate controls in each application consolidated into TrustGate, which applies shared controls and routes several applications to multiple LLMs.

What you can do

Route across providers

Fallback chains, load balancing, and smart routing by request complexity.

Govern traffic

Rate limits, LLM budgets, tool allow-lists, prompt controls, and guardrails.

Speak one API

Chat, embeddings, images, files, audio, rerank, and model discovery.

See supported providers

Every adapter TrustGate can route to, and which modalities it supports.

Expose MCP tools

One MCP gateway for Claude Code, Cursor, Codex, Copilot, and Copilot Studio.

Authenticate consumers

API keys, or JWTs validated against your IdP.
Attach TrustGuard when you want inline inspection of prompts, responses, and tool activity. See the integrations catalog for coding agents and other gateways.

Core objects

Everything in the console under TrustGate is one of these:

Request path

The client names a model or uses auto. It does not send a provider URL or key. Architecture covers the full lifecycle.

FAQ

Yes. Change the base URL and replace the provider credential with a TrustGate consumer credential. TrustGate accepts OpenAI Chat Completions, Responses, Anthropic Messages, and Google Gemini formats, then adapts the request for the selected upstream. Start from the connection snippet on the consumer Connect tab.
Use an API key for a static application. Or send a JWT — TrustGate validates it against your IdP and can route from token claims. See Authentication.
Yes. Hybrid keeps the data plane in your network. External runs the full platform yourself. SaaS hosts both planes. Compare models in Deployment.
No. Provider credentials stay on the registry. Applications authenticate as a consumer. Rotate consumer keys separately from provider keys.