wickedailabs-agentframework 0.2.0a1

Scaffolding framework over the Microsoft Agent Framework for Python.

wickedailabs-agentframework

Scaffolding framework over the Microsoft Agent Framework for Python.

wickedailabs-agentframework owns all infrastructure — dependency injection, settings, provider wiring, tool discovery, runners (simple / HITL / workflow / orchestration), session storage, middleware, OpenTelemetry tracing and metrics. Your agent projects contain only domain logic: @agent classes, @agent_tool methods, and @service implementations.

Install

This package is distributed via GitHub Releases (not PyPI). Pick the approach that fits your project:

From a release wheel:

pip install https://github.com/WickedAILabs/wickedailabs-pyagentframework/releases/download/v0.2.0a1/wickedailabs_agentframework-0.2.0a1-py3-none-any.whl

From a git tag (subdirectory-install of this package):

pip install "git+https://github.com/WickedAILabs/wickedailabs-pyagentframework.git@v0.2.0a1#subdirectory=src/wickedailabs_agentframework"

In pyproject.toml:

[project]
dependencies = [
  "wickedailabs-agentframework @ https://github.com/WickedAILabs/wickedailabs-pyagentframework/releases/download/v0.2.0a1/wickedailabs_agentframework-0.2.0a1-py3-none-any.whl",
]

Optional extras apply the same way — e.g. "wickedailabs-agentframework[foundry] @ <url>". Available extras: foundry, azure-ai, copilot-studio, ollama, yaml, all.

Minimal usage

import asyncio
import my_agent_project
from wickedailabs_agentframework import (
    AgentAI, AgentClient, AgentContext,
    add_agent_framework, build_settings,
    agent, agent_instructions, uses_tools,
)
from wickedailabs_agentframework.internal.runners import (
    SimpleRunner, HitlRunner, WorkflowRunner, OrchestrationRunner,
)


@agent
@agent_instructions("You are a helpful weather advisor.")
@uses_tools("get_current_weather")
class WeatherAdvisorAgent:
    async def run(self, message: str, ctx: AgentContext) -> str:
        return await AgentAI.ask(self, message)


async def main() -> None:
    settings = build_settings()
    container = add_agent_framework(
        settings,
        runners=[SimpleRunner(), HitlRunner(), WorkflowRunner(), OrchestrationRunner()],
        caller_package=my_agent_project,
    )
    client = container.resolve(AgentClient)
    result = await client.run("WeatherAdvisor", "What's the forecast in Sydney?")
    print(result.response if result.success else f"[{result.error_code}] {result.error}")


asyncio.run(main())

Configuration lives in appsettings.json (or .toml / .yaml), with environment-variable overrides via the AgentFramework__Provider__Type=... convention.

Links

• GitHub repo — full spec (spec.md), sample projects/weather_advisor/, and contribution guide.
• Releases — download wheels + sdists.
• CHANGELOG
• Issues

Per-agent tool options

Agents can opt into provider-hosted or external tools via config:

"BudgetAdvisor": {
  "Runner": "simple",
  "Instructions": "You reconcile budget variances.",
  "InstructionsFile": "prompts/budget.txt",           // alternative to Instructions
  "McpServers": [
    { "Name": "reports", "Url": "http://localhost:8000/mcp" }
  ],
  "CodeInterpreter": true,
  "FileSearch": { "VectorStoreIds": ["vs_policies"] }
}

At call time, attach local files via FileAttachment.from_path / from_bytes / from_uri. Capability enforcement at startup rejects combinations the provider can't support (HITL on Anthropic, MCP on Copilot Studio, etc.). MCP URLs and from_uri both require http/https.

License

GPL-3.0-or-later — see the LICENSE file in the repository.