Quick Starts

Choose the runtime or language you already use. Relay attaches where your agent, framework, or application owns model requests and tool execution.

Relay has no required or default model. Each guide that makes live inference requests identifies the credentials and model access it needs. The application API examples below use local callbacks.

Agent Harnesses

When a coding-agent runtime owns invocation, start with CLI Basic Usage. For persistent Claude Code or Codex installation, use Coding Agent Installation. For agent-owned integrations, refer to Supported Integrations.

Codex

Run one Codex CLI task with Relay model and tool observability.

Claude Code

Run one Claude Code session with Relay model and tool observability.

Hermes Agent

Configure Hermes Agent’s native Relay observability and run one agent turn.

Install the proof-of-concept extension and observe one pi task.

Featured Plugins

For reusable runtime behavior, start with Plugin Configuration Files.

Agent Frameworks

If a framework owns callbacks or tool execution, choose its path in Supported Integrations.

Deep Agents

Run one Deep Agents workflow with Relay’s maintained integration.

LangChain

Run one tool-using agent with Relay middleware and callbacks.

LangGraph

Run one credential-free graph with Relay callbacks.

Application APIs

Use these paths when your application directly owns the tool or model call.

Python

Instrument application-owned Python calls.

Node.js

Instrument application-owned Node.js calls.

Rust

Instrument application-owned Rust calls.

For More Details

For installation and exporter options, refer to: