Open standard for declarative AI agent systems.
Define your agents, tools, and multi-agent orchestration in portable AFS files. Run them on any compliant runtime.
AI agent configurations are fragmented. Every framework has its own format, locking you into a specific runtime. Agent Formation changes that:
- Write once - Declarative schemas for agents, MCP tools, and A2A services
- Run anywhere - Portable across any compliant runtime
- Validate consistently - Standard tooling for linting and validation
- Extend safely - Namespaced extensions for vendor-specific features
Think of it as infrastructure-as-code for AI agents.
A Formation is a complete AI agent system defined in AFS (or YAML):
# formation.afs
schema: "1.0.0"
id: my-formation
description: Customer support system
llm:
models:
- text: openai/gpt-4o
agents:
- id: support-agent
role: Customer Support
tools: [knowledge-base, ticketing]
mcp:
- id: knowledge-base
command: npx @company/kb-serverAgents, tools, memory, knowledge, workflows - all declared, all portable.
File Extensions: Use
.afs(Agent Formation Schema) or.yaml- both are fully supported.
| Repository | Description |
|---|---|
| afs-spec | Core schemas, specs, and templates |
| afs-cli | Validator, linter, and formatter tool |
| afs-vscode | VS Code extension |
| afs-assets | Logos, icons, badges, and brand guidelines |
# Clone the formation templates
git clone https://github.com/agent-formation/afs-spec.git
cp -r afs-spec/schemas my-project
cd my-project
# Customize and deploy with any compliant runtime- MUXI - Reference implementation
Building a runtime? We'd love to list it. Open an issue or PR.
Agent Formation is developed in the open. We welcome contributions to schemas, specs, tooling, and documentation.
- Read the Contributing Guide
- Review the Governance Model
- Join the discussion in Issues
Apache License 2.0
Portable agents. Open standard. Community driven.