Provider-native agents for reusable workflows

Minions provides one consistent workflow-agent family across OpenAI, Google Gemini, xAI Grok, Anthropic Claude, and Mistral. Each provider module accepts its own native models, tools, return types, and streaming interface—without translating provider contracts or forcing tools into workflows that do not need them.

Get started Browse the API

---

Where Minions fits

Package	Responsibility
Guro	Reusable system instructions organized by workflow category
Fonky	Reusable provider-compatible functions and tool schemas
Minions	Provider agents, native tools, tool execution, async execution, and streaming

Provider pathways

Module	Provider SDK	Implementation	Native execution
minions.gpt	OpenAI Agents SDK	Inherits agents.Agent	run, run_async, stream
minions.gemini	Google ADK	Inherits google.adk.Agent	run, run_async, stream
minions.grok	xAI SDK	Owns sync and async chats	run, run_async, stream
minions.claude	Anthropic SDK	Owns native tool runners	run, run_async, stream
minions.mistral	Mistral SDK	Owns a remote agent	run, run_async, stream

Workflow templates

Every provider exposes the same fourteen concrete workflow roles:

ResearchMinion, WritingMinion, ComplianceMinion, BusinessMinion, CodingMinion, DataMinion, GovernanceMinion, PlanningMinion, ImageGenerationMinion, ImageAnalysisMinion, ImageEditingMinion, TranslationMinion, TranscriptionMinion, and SpeechMinion.

The class communicates intent. The instructions, model, and optional tools determine the actual behavior.

Design guarantees

• Provider fidelity — tools and response objects stay native to the selected SDK.
• Optional tools — tool-free workflows remain first-class.
• Complete tool loops — local calls continue until the provider returns a final response.
• Explicit limits — max_turns guards every workflow against an unbounded loop.
• Typed public surface — provider-specific result and tool types remain visible.

Explore the architecture