PRISM

AI-Native Autonomous Materials Discovery Platform

by MARC27

---

PRISM is a single binary that gives you an AI agent for materials science. It searches knowledge graphs, runs compute jobs, orchestrates workflows, and connects to a federated mesh of research nodes.

prism                    # Launch interactive chat
prism query --platform "nickel superalloys"
prism research "high-entropy alloys" --depth 1
prism billing            # Check credit balance

Install

macOS / Linux

curl -fsSL https://prism.marc27.com/install.sh | bash

Windows (PowerShell — not Command Prompt)

irm https://prism.marc27.com/install.ps1 | iex

The installer runs in two stages: it puts the binary on your PATH, then triggers prism doctor, which provisions the Python tool platform and prints what is and isn't ready. Re-running it resumes rather than starting over.

Requirements

Python 3.11 or newer	Required, not optional. The agent runs its tools in a Python worker; without 3.11+ prism exits immediately with No Python 3.11+ found. macOS's built-in python3 is 3.9 — brew install python@3.12.
Linux: glibc 2.35+	Ubuntu 22.04+, Debian 12+, SLES 15 SP6+. RHEL/Rocky 9 ship glibc 2.34 and are not supported by the prebuilt binaries — build from source there.
Disk	~250 MB for the binary and ~400 MB for the Python environment.

Supported platforms

Platform	Archive	Status
macOS Apple Silicon	prism-macos-aarch64.tar.gz	Supported, macOS 11+
macOS Intel	prism-macos-x86_64.tar.gz	Supported; no local embedding model — see below
Linux x86_64	prism-linux-x86_64.tar.gz	Supported, glibc 2.35+
Linux ARM64	prism-linux-aarch64.tar.gz	Supported, glibc 2.35+
Windows x86_64	prism-windows-x86_64.zip	Experimental — see below

Windows is experimental. The installer and the binary are there, and ARM64 Windows works under x64 emulation, but PRISM resolves its data directory from HOME, which stock Windows does not set, in several places outside the install path. Expect rough edges beyond prism --version. If you need PRISM working today, use macOS or Linux — including WSL2, which is a fully supported Linux target.

Downloading the archive directly

Use the installer if you can — it handles the platform quirks below. If you download from GitHub Releases by hand:

macOS. The binaries are ad-hoc signed but not notarized with an Apple Developer ID, so Gatekeeper blocks them with "Apple could not verify this app is free of malware". Clear the quarantine flag after extracting:

tar -xzf prism-macos-aarch64.tar.gz
xattr -d com.apple.quarantine prism prism-node
./prism --version

Windows. The .exe is unsigned. Launched from PowerShell or Command Prompt it runs without a prompt; double-clicking it in Explorer shows a SmartScreen "Windows protected your PC" warning, where you would click More info → Run anyway.

Linux. Extract and run — no signing involved.

From source

Needs protoc and Node 22 (for the bundled dashboard):

git clone https://github.com/Darth-Hidious/PRISM.git
cd PRISM && (cd dashboard && npm ci && npm run build)
cargo build --release --bin prism --bin prism-node
cp target/release/prism target/release/prism-node ~/.local/bin/

Known platform limits

• Intel Macs have no local embedding model: ONNX Runtime publishes no x86_64-apple-darwin build, so semantic search falls back to keyword-only unless you point PRISM_EMBED_BACKEND=openai at an embeddings endpoint. Everything else is identical.
• Homebrew is not currently a supported install route.

Interactive chat

Run prism to launch a slash-command-driven AI agent with:

• Slash-command palette — type / for commands like /new, /info, /usage, /help, /model, /agent, /update, /exit
• Model picker via /model (hundreds of hosted models routed through MARC27)
• Inline tool cards showing materials-science tool execution results (CALPHAD, pyiron, OPTIMADE, ML predictions, federated knowledge graph)
• Streaming AI responses with markdown rendering
• Boot diagnostics (Platform, Auth, Knowledge Graph, LLM Models, Compute, Marketplace, Local Node, Policy Engine) on every launch — run prism doctor for the full report

CLI Commands

Setup & Auth

prism setup              # First-time setup + login
prism login              # Authenticate with MARC27
prism configure --show   # Show LLM config
prism status             # Auth state, paths, endpoints

AI Agent

prism                    # Interactive chat agent
prism query --platform "titanium alloys"
prism query --semantic "creep resistance"
prism query "Ti-6Al-4V"                  # Local knowledge-graph lookup
prism research "novel refractory alloys" --depth 1

Knowledge Graph

prism ingest ./data.csv              # Ingest data
prism ingest --schema-only ./data.csv
prism ingest --watch ./data/

Compute

prism run python:3.11 --input data=x.csv
prism run --backend marc27 --gpu A100
prism run --ssh user@host             # BYOC via SSH
prism run --k8s-context prod          # BYOC via Kubernetes
prism run --slurm head@cluster        # BYOC via SLURM
prism deploy create --name my-service --image img:latest
prism deploy list
prism job-status <uuid>

Mesh & Nodes

prism node up            # Start local node (bundled knowledge graph)
prism node status
prism mesh discover      # Find LAN peers via mDNS
prism mesh publish       # Share dataset to mesh

Models

prism models list        # 535 hosted LLMs
prism models search "claude"

Workflows

prism workflow list
prism workflow run explore --space "Ni-Cr-Co" --target "yield_strength > 900"

Billing

prism billing            # Credit balance
prism billing usage      # Usage breakdown
prism billing topup      # Buy credits (Stripe)
prism billing prices     # Pricing table

Other

prism tools              # List 123 available tools
prism marketplace search # Browse datasets, models, plugins
prism discourse list     # Multi-agent debate workflows
prism report "bug"       # File support ticket

Architecture

prism (single Rust binary)
  prism-cli      Command routing, auth, TUI
  prism-llm      LLM client (OpenAI + MARC27 proxy)
  prism-agent    TAOR agent loop, tool calling, OPA policy
  prism-node     Daemon, probe, E2EE key exchange
  prism-ingest   Schema detection, ontology extraction
  prism-server   Axum REST API + dashboard
  prism-mesh     mDNS + Kafka pub/sub + federation
  prism-compute  Docker / MARC27 / SSH / K8s / SLURM
  prism-policy   OPA/Rego policy engine
  prism-workflows YAML workflow engine (8 step types)

Python tools (123 tools via JSON stdio)
  Search: OPTIMADE (20+ providers), arXiv, Semantic Scholar
  Predict: property prediction, structure prediction
  Simulate: CALPHAD, DFT planning
  Execute: Python, Bash (sandboxed)
  Platform: knowledge graph, compute, marketplace
  Custom: drop .py in ~/.prism/tools/

Configuration

All config in ~/.prism/prism.toml:

[llm]
provider = "marc27"
url = "https://api.marc27.com/api/v1/projects/{id}/llm"
model = "gemini-3.1-flash-lite-preview"

[node]
name = "lab-alpha"
port = 7327

[mesh]
discovery = ["mdns", "platform"]

Security

• OPA/Rego policy engine — every tool call checked
• E2EE — X25519 + ChaCha20-Poly1305 for node-to-node data
• RBAC — 4-tier roles (admin, operator, agent, viewer)
• Audit logging — every action logged

Contributing

• TUI — built with Ratatui, protocol documented in docs/FRONTEND_PROTOCOL.md
• Tools — new Python tools in app/tools/, custom tools in ~/.prism/tools/
• Workflows — YAML in ~/.prism/workflows/, auto-discovered as CLI commands

License

Component	License
Python tools, tool server, config, plugins	MIT
Rust crates, TUI, agent, workflows	MARC27 Source-Available

Commercial licensing: team@marc27.com

---