AgentHealth Python SDK

Typed synchronous and asynchronous access to AgentHealth's Go CLI. This package preserves AHS v1 results, diagnostic codes, dependency relationships and graph IDs. Health failures return results; tool failures raise exceptions.

Install the package with pip install agenthealth once published. Install a verified AgentHealth CLI of the same software version separately and put it on PATH, or pass its absolute path. Installation and imports never download or execute a binary. This first SDK targets Python 3.10+ and CLI v0.17.0.

from agenthealth import Client, HealthStatus

client = Client(binary="/usr/local/bin/agenthealth", timeout=30)
report = client.check(target_type="http", endpoint="http://localhost:8080/health")
print(report.status, report.exit_code)
if report.status == HealthStatus.HEALTHY:
    print(report.result.target.name)

from agenthealth import async_check

async def probe():
    return await async_check("agenthealth.yaml")

Use Configuration for Python dictionaries or load_config for JSON/YAML. The same configuration supports all implemented CLI adapters, explicit active checks, nested dependencies and graph references. Credentials stay in configured references. Configuration is trusted executable input: MCP stdio can launch programs and endpoints can reach private networks. Loading validates schema; Go validates graph semantics at execution.

The SDK checks the exact CLI version, bounds configuration and output, enforces an overall deadline and stops checks on timeout/cancellation. POSIX checks use a dedicated process group to clean up descendants. On Windows termination covers the CLI process; detached stdio descendants require operator lifecycle controls. CLI stderr is never included in exceptions.

The SDK does not aggregate health independently, run OAuth login, register Python adapters, or auto-install the CLI. report.results preserves ordered roots, including each dependency projection. to_dict() preserves additional result fields. Relative paths retain the caller's working directory, matching the CLI; Configuration(data, base_dir=...) selects a different directory.

Source implementation is pre-1.0; PyPI availability is recorded separately in the repository capability status. See the repository's Python SDK guide for installation, release setup, errors and examples.

This package version: v0.17.0. Requires the matching CLI; model/LLM configuration is supported through the canonical schema.

Native database targets use the matching CLI; fixed SQL queries require explicit functional opt-in. Passwords use database.password_env references.

Qdrant/Pinecone vector-store checks are included in this version; use the matching v0.17.0 CLI. Other vector providers remain planned.