AdaL Agent SDK
The AdaL Agent SDK is a Python package that lets you embed AdaL's full agent runtime in your own applications. Instead of using the CLI interactively, you can drive AdaL programmatically — run queries, stream events, handle tool approvals, resume sessions, and orchestrate multi-step workflows from Python.
Use Cases
- CI/CD pipelines — run code review, test generation, or refactoring as automated steps
- Custom tooling — build internal developer tools powered by AdaL's agent capabilities
- Batch processing — process multiple files or repositories in sequence
- Testing & evaluation — script reproducible agent interactions for benchmarking
- Integrations — embed AdaL in Slack bots, web services, or notebook workflows
Architecture
Your Python App
│
├─ query() ← one-shot convenience API
│
└─ AdalAgentClient ← persistent multi-query client
│
▼
AdaL SDK Runtime ← subprocess (adal --sdk-runtime)
│
▼
AdaL Backend ← agent logic, tools, LLM calls
The SDK spawns AdaL as a subprocess and communicates over a stdio NDJSON protocol. This means:
- Full agent capabilities — same tools, models, and reasoning as the CLI
- Session persistence — resume conversations across client instances
- Streaming events — real-time access to assistant text, tool calls, and status updates
- Permission control — programmatic approve/deny/modify for any tool call
Requirements
| Requirement | Version |
|---|---|
| Python | 3.10+ |
| AdaL CLI | Latest (with --sdk-runtime support) |
| Dependencies | anyio>=4.0 |
The SDK requires the AdaL CLI installed and authenticated. Install it with:
# macOS, Linux, WSL
curl -fsSL https://adal.sylph.ai/install.sh | bash
# Windows PowerShell
irm https://adal.sylph.ai/install/windows | iex
Run adal once to complete browser authentication before using the SDK.
What's Next
- Quick Start — install the SDK and run your first query
- Client API — multi-query sessions, resume, and model switching
- Permissions — control tool approval with callbacks
- API Reference — full reference for all exports