Skip to main content
Claude Desktop ships with a setting called third-party inference. When enabled, every conversation goes to a gateway you choose instead of Anthropic’s hosted API. Pointing it at Portkey turns a personal AI tool into a governed organizational one without changing how anyone uses the app.

​
What changes when you connect Portkey

One endpoint, every model

Use Anthropic, Bedrock, or Vertex AI without changing anything in Claude Desktop. Switch providers from a config in Portkey.

Spend stays within limits

Per-team budgets, rate limits, and content guardrails apply automatically. No more surprise invoices.

Every request is auditable

Cost, latency, tokens, and metadata logged for every prompt. Filter by team, user, or environment.

​
Pick your path

I'm rolling this out for my team

Platform Admin Guide. Set up providers, configs, governance policies, and scoped API keys. Hand each developer a five-minute setup.

I just need to connect Claude Desktop

Developer Guide. Paste the credentials your admin gave you and start sending requests through the governed gateway.
Solo builder or pilot? Run through the Platform Admin Guide once to create your provider and config. Then follow the Developer Guide to wire up Claude Desktop. Total time: under fifteen minutes.

​
When to use this guide vs. the others

Claude Desktop

The standard chat experience in Claude Desktop. You’re here.

Claude Cowork

The collaborative coding experience inside Claude Desktop. Same gateway setup, slightly different surface.

Claude Code

The CLI-based agentic coding tool. Different config format (settings.json).
Last modified on June 3, 2026