Agent Configuration Overview

Agent Configuration lets you manage your prompts, model settings, and tool definitions outside your codebase. Version them, update them without redeploying, and keep a full history of every change.

Why use Agent Configuration

When prompts and parameters are hardcoded, every change requires a deploy. There is no history of what changed, no way to roll back quickly, and no clean way to test a different prompt in staging vs. production.

Agent Configuration gives you version control for your agent's parameters. Define them once in code, push them to Opik, and update, compare, and promote versions through the UI or SDK.

Key features

• Version control — Every change creates an immutable version. You can always see what changed and roll back.
• Environment labels — Tag a version as prod or staging in the UI. Your agent picks it up automatically.
• Prompt management — Version text prompts and chat prompts, with {{variable}} substitution built in.
• Typed parameters — Store model names, temperatures, tool definitions, and any other string or numeric values alongside your prompts.
• Python and TypeScript SDKs — Define and fetch configs from either language.

How it works

1. Define your configuration — Declare a Config subclass (Python) or a plain object (TypeScript) with your fields and defaults, then push it with create_config / createConfig.
2. Fetch it at runtime — Inside a tracked function, call get_or_create_config / getOrCreateConfig. The SDK returns the version that matches your requested environment or version. If Opik is temporarily unreachable, it falls back to the defaults you provided.

When you want to change something, create a new version and move the prod label to it. Your running agent picks up the change on its next fetch — no redeploy needed.

Next steps

• Getting started — Add Agent Configuration to your code in minutes
• Concepts — Deep dive into versions, environments, and configuration types