Extending the Agent
The integration surfaces change how you invoke AdaL. What the agent can do is customized separately — and that customization applies identically to interactive, headless, and SDK runs. Build the tool or prompt once, and every surface benefits.
Where to customize
The Customize AdaL section covers everything:
- Custom Tools — teach the agent new capabilities in
.adal/tools.py - Skills & Plugins — reusable domain expertise, installable from marketplaces
- Custom System Prompt —
AGENTS.mdand--prompt-file - Lifecycle Hooks — your logic at key points in the agent lifecycle
- Model providers — your own API keys, endpoints, or local models
Putting it together
A typical third-party integration combines layers: headless mode or the SDK to drive AdaL, custom tools to give it domain-specific abilities, and an AGENTS.md to shape its behavior for your context. When a configuration is worth sharing or running unattended, package it as a Cloud Agent — see Cloud Agents.