Skip to content
MARC27

Introduction

MARC27 is a platform for materials science research. It connects a curated knowledge graph, 519 LLM models, multi-agent discourse, compute brokerage, and a research engine into one system.

What You Get

Section titled “What You Get”
  • Knowledge Graph — 69,000+ entities and 5.3M+ relationships curated from 5 million scientific papers, plus 9,000+ paper content embeddings for semantic search
  • Research Engine (RLM) — Ask questions, get answers grounded in real graph data. The LLM writes code to explore the graph iteratively — no hallucination
  • 519 LLM Models — Anthropic, OpenAI, Google, and OpenRouter. Bring your own keys or use platform keys
  • Multi-Agent Discourse (MiroFish) — Define agent debates in YAML. Agents challenge claims, reach consensus, run A/B tests
  • Compute — Deploy models from HuggingFace to GPUs (RunPod, AWS, Lambda). Budget caps and auto-stop
  • PRISM Nodes — Run compute on your own hardware. E2E encrypted data transfer between nodes

Architecture

Section titled “Architecture”
Layer 4: Spark     — Project orchestration (months, phase gates)
Layer 3: MiroFish  — Multi-agent discourse (YAML workflows)
Layer 2: RLM       — Research engine (iterative REPL)
Layer 1: CFE       — Scoring kernel (relevance ranking)

Everything routes through a single API gateway at api.marc27.com. GraphQL is the primary interface — 20 queries and 6 mutations covering every service.

For Enterprise

Section titled “For Enterprise”
  • Tenant isolation — your data never mixes with others
  • On-premises — run the full stack on your infrastructure with prism node up
  • E2E encryption — X25519 + AES-256-GCM between nodes
  • Platform identity — every user gets username@id.marc27.com as a verifiable signature

Getting Started

Section titled “Getting Started”
  1. Quick Start — install PRISM and run your first query
  2. Authentication — API keys, device flow, JWT
  3. API Reference — REST and GraphQL endpoints