MUSE

Modular Stack of Evidence — An evidence-based impact planning tool for Digital Public Goods.

Overview

MUSE helps organizations plan and measure social impact using the Theory of Change methodology. It combines AI-powered logic model generation with research evidence validation and blockchain-based attestations to create transparent, evidence-backed impact pathways.

Built as part of the Beacon Labs ecosystem for supporting evidence-based practice (EBP) in Digital Public Goods (DPG).

Features

AI-Powered Logic Models

Mastra-based AI agents generate complete Theory of Change logic models through a 5-stage process: analyze context, generate structure, design visual layout, self-critique, and produce canvas-ready output. The result is a fully connected pathway from Activities → Outputs → Short-term Outcomes → Intermediate Outcomes → Impact.

Evidence-Based Validation

An LLM-powered evidence search agent semantically matches research evidence against every causal relationship in a logic model. Using batch processing and chain-of-thought reasoning, it identifies which connections are backed by published research — making the distinction between evidence-supported and theoretical pathways clear.

Blockchain Attestation

Evidence submissions are attested on-chain via EAS (Ethereum Attestation Service) on Base Sepolia, with content stored on IPFS. Logic models can generate Hypercerts for transparent impact tracking and measurement.

Interactive Canvas

A React Flow-powered visual builder for creating and editing logic models. Evidence-backed edges are highlighted in green, and each edge includes an interactive button to view the supporting research details, scores, and methodology.

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                         MUSE Platform                           │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│  Evidence Repository          MUSE Application                  │
│  ┌──────────────┐            ┌──────────────────────────────┐  │
│  │ MDX Research  │  npm pkg  │                              │  │
│  │ Files         ├──────────►│  AI Agents (Mastra)          │  │
│  │              │            │    ├─ Logic Model Agent      │  │
│  │ Zod          │            │    └─ Evidence Search Agent  │  │
│  │ Validation   │            │           │                  │  │
│  └──────┬───────┘            │           ▼                  │  │
│         │                    │  React Flow Canvas           │  │
│         │ GitHub Actions     │    ├─ Visual Logic Models    │  │
│         ▼                    │    └─ Evidence-backed Edges  │  │
│  ┌──────────────┐            │           │                  │  │
│  │ IPFS + EAS   │            │           ▼                  │  │
│  │ Attestation  │            │  Hypercerts (Impact)         │  │
│  └──────────────┘            └──────────────────────────────┘  │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

Getting Started

Prerequisites

• Node.js 20+
• Bun 1.3+

Installation

git clone https://github.com/beaconlabs-io/muse.git
cd muse
bun install

Environment Setup

Copy the example environment file and fill in the required values:

cp .env.example .env.local

Key variables include API keys for AI providers, PINATA_JWT for IPFS uploads, and NEXT_PUBLIC_WALLETCONNECT_PROJECT_ID for wallet connection. See .env.example for the full list.

Development

bun dev

Open http://localhost:3000 to see the application.

Scripts

Command	Description
bun dev	Start Next.js development server
bun run build	Build for production
bun start	Start production server
bun lint	Run ESLint with auto-fix
bun format	Format code with Prettier
bun clean	Clean build artifacts and reinstall
bun dev:mastra	Start Mastra agent development server
bun build:mastra	Build Mastra agent system

Project Structure

.
├── app/                  # Next.js App Router
│   ├── canvas/           #   Interactive logic model builder
│   ├── evidence/         #   Evidence browsing and detail pages
│   ├── hypercerts/       #   Hypercerts integration
│   └── api/              #   Server-side API endpoints
├── components/           # React components
│   ├── canvas/           #   React Flow nodes, edges, and controls
│   ├── evidence/         #   Evidence-specific UI components
│   └── ui/               #   shadcn/ui primitives (auto-generated)
├── mastra/               # AI agent system
│   ├── agents/           #   Logic model and evidence search agents
│   ├── workflows/        #   Multi-step agent workflows
│   ├── tools/            #   Agent tools (canvas data, evidence access)
│   └── skills/           #   Domain knowledge for agents
├── lib/                  # Shared utilities and configuration
├── hooks/                # Custom React hooks
├── types/                # TypeScript type definitions
└── docs/                 # Technical documentation

Documentation

For detailed technical information, see:

Document	Description
AI Agent Architecture	Agent workflows, quality controls, Skills API
Evidence Workflow	Submission, attestation, search philosophy
React Flow Architecture	Canvas implementation, evidence edges, UI flow

Deployments

	Environment	URL
Production	production	https://muse.beaconlabs.io
Development	development	https://dev.muse.beaconlabs.io

Contributing

Contributions are welcome! Please see CONTRIBUTING.md for development workflow, code style guidelines, and the pull request process.

Evidence files are managed in a separate repository: beaconlabs-io/evidence.

License

This project is licensed under the Apache License 2.0.