Skip to main content
When developers use Claude Code or Codex individually, it works fine. When 100 developers use it — raw API keys get scattered, costs spike overnight without attribution, and there’s no fallback when a provider goes down. Portkey is the gateway that sits between your team’s coding agents and LLM providers. Developers change nothing. You get full control.

Secure

Centralize provider credentials. Issue scoped API keys per team. Rotate or revoke access without touching developer configs.

Control

Set hard budget and rate limits per developer, team, or workspace. Block requests when limits are hit.

Observe

Log every request — cost, tokens, latency, model, and who made it. Attribute spend by team or project.

​
Supported agents

Claude Code

The most widely deployed AI coding agent. Supports Anthropic direct, Bedrock, and Vertex AI through a single Portkey endpoint.

OpenAI Codex

OpenAI’s terminal coding agent. Configure via config.toml with full provider flexibility.

Cursor

Route Cursor’s AI requests through Portkey for observability and cost controls.

Cline

Set Portkey as Cline’s custom OpenAI base URL.

Roo Code

Drop-in Portkey integration via base URL override.

Pi Agent

Minimal extensible terminal coding agent. Route through Portkey via custom provider config.

Goose, opencode & more

Any OpenAI-compatible coding agent works with Portkey.

​
What platform teams get

​
Access control

Developers never see provider credentials. Every request is authenticated through a scoped Portkey key.

​
Cost management

Agentic loops are expensive. One uncapped session can rack up hundreds of dollars overnight.
Per-request cost breakdown and metadata filters in Portkey

​
Reliability

Coding agents fail silently when providers go down. Portkey routes around outages automatically.

​
Governance

Coding agents operate with terminal-level access. Portkey adds the control layer.

​
Setting up Claude Code and Codex

For Claude Code and Codex, the Portkey CLI sets up gateway routing, MCP servers, and team skills in one command:

Claude Code setup

Anthropic direct, Bedrock, and Vertex AI

Codex setup

Full config.toml reference and provider options

​
Next steps

Budget Policies

Hard spend limits per team or developer

Model Catalog

Centralized provider credentials and access controls

Configs

Fallbacks, load balancing, and routing strategies

MCP Gateway

Govern MCP tool access across your agent fleet

Observability

Logs, traces, and cost breakdowns for every session

Agent Skills

Sync team coding standards to every developer’s agent
Last modified on April 21, 2026