Customize AdaL
Most agents are fixed at install. AdaL is built to be shaped: teach it new tools, give it domain expertise, rewrite its prompt, and run your own models — customize once and it applies everywhere, from your interactive session to CI runs to the SDK.
Personalize the agent
- Custom System Prompt — give the agent project-specific instructions with an
AGENTS.mdfile, or replace its role prompt entirely with--prompt-file. - Lifecycle Hooks — run your own commands at key points in the agent lifecycle: enforce guardrails, inject context, trigger local workflows.
Teach it new abilities
- Custom Tools — write Python functions in
.adal/tools.pyand the agent can call them like built-in tools. - Skills & Plugins — install reusable domain expertise from community marketplaces (compatible with Claude Code skills), or package your own for your team.
Choose your models
- Bring Your Own API Key — use your own provider keys, optionally pointed at your own gateway or self-hosted endpoint.
- ChatGPT Subscription — connect a ChatGPT Plus or Pro subscription to use Codex models without an API key.
- Local Models — run open-source models with Ollama, fully offline.
Combining with integrations
Customization composes with the integration surfaces: a headless run in CI uses the same custom tools and AGENTS.md as your interactive session, and a Cloud Agent wraps a system prompt plus custom tools into a reusable, shareable configuration — see Cloud Agents.