fakzed

A lightweight Python framework for building agent platforms.

• Three primitives: Agent (tool loop + multi-turn), Team (coordinate / route / collaborate), Workflow (sequential / conditional / loop / parallel)
• Own model layer: OpenAI / DeepSeek / Anthropic / Ollama adapters with a registry factory ("deepseek:deepseek-chat"), streaming, tool calls, reasoning content, structured output
• Pluggable capabilities: @tool + built-ins + MCP client, AgenticMemory, Knowledge RAG, Skills, ContextProvider, Guardrails, Hooks, observability
• Runtime: BackgroundRunner, Scheduler, Evals

uv add fakzed            # or: pip install fakzed
uv add fakzed[deepseek]  # provider extras: openai / anthropic / ollama / deepseek / sqlite

from fakzed import Agent, tool


@tool
def add(a: int, b: int) -> int:
    """Add two numbers."""
    return a + b


agent = Agent(name="calc", model="deepseek:deepseek-chat", tools=[add])
resp = agent.run("1 + 1 = ?")
print(resp.content)

See the monorepo README, docs and cookbook for the full guide.