Leverbrain

The expertise registry for the agentic era.

┌────────────────────────────────────────────────────────────────────────┐
│  $ npm install -g leverbrain                                           │
│  $ leverbrain search deep-research                                     │
│  $ leverbrain get leverbrain/leverbrain                                │
│                                                                        │
└────────────────────────────────────────────────────────────────────────┘

The Prompt Wrapper Fallacy

When you watch people build AI agents today, you notice a recurring pattern. Most builders are setting up wrappers around generic prompts. An agent is instructed: "You are a world-class copywriter. Write a viral launch campaign."

The model does its best, but its best is a collection of commonplaces. It has never actually launched a product. It doesn't know the physical constraints of the X recommendation engine, how Kafka streams rank replies, or how to design hooks for 3-second retention cliffs.

An agent shouldn't have to guess how to perform a task. It should ingest a verified blueprint authored by a human who has already done it.

Leverbrain is a high-fidelity registry and marketplace for AI agent skills, strategies, and blueprints. Instead of telling your agent what to be, you tell it what to load.

How it Works

The architecture is simple and local-first:

       ┌──────────────┐         ┌──────────────┐
       │   Developer  │ ──────> │  Solana (L1) │ (Verified Receipts)
       └──────────────┘         └──────────────┘
              │
              ▼ (Publish / Buy via CLI)
       ┌──────────────┐         ┌──────────────┐
       │  Leverbrain  │ <─────> │   Convex DB  │ (Real-Time Registry)
       │  Marketplace │         └──────────────┘
       └──────────────┘
              │
              ▼ (Local Sync)
       ┌──────────────┐
       │ Local Agent  │ ──> [Ingests SKILL.md & runs local shims]
       └──────────────┘

1. Sovereign BLUEPRINTS (`SKILL.md`): Skills are defined in standard markdown files with structured YAML frontmatter (defining constraints, taglines, categories, and execution rules).
2. On-Chain Licensing (Solana): Purchases are recorded as on-chain receipts (PDAs) on the Solana Mainnet. Devs monetize their execution files, and buyers have cryptographically verified ownership.
3. Low-Latency Sync (Convex): The Convex database acts as the high-speed registry, serving verified TypeScript modules and markdown guides directly to local terminal agent runtimes (like Claude Code, Cursor, or custom frameworks).

Developer Quickstart

Skip the web browser. Keep your hands on the keyboard.

1. Install CLI

Run directly without installing:

npx --yes leverbrain@latest --help

2. Search & Get Skills

Browse the registry and download a package directly into your workspace:

leverbrain search x-algo
leverbrain get santa/x-algo

3. Deploy Configuration

Load a unified agent workspace setup:

leverbrain cfg name/cfg

CLI Command Reference

leverbrain search <query>           # Search the marketplace
leverbrain get <author/slug>        # Download a purchased skill package
leverbrain cfg <name/cfg>        # Download a saved configuration
leverbrain purchases --wallet <pk>  # List receipts for a Solana wallet
leverbrain publish ./my-skill \
  --wallet <KEYPAIR_PATH> \
  --author <HANDLE>                 # Sign & publish a new SKILL.md

Authentication

Protected actions (downloads and publishing) require Solana signature authentication. The CLI automatically signs a short-lived message payload and attaches the following headers:

X-Wallet-Address: <base58 public key>
X-Wallet-Signature: <base58 signature of message>
X-Wallet-Message: leverbrain-auth-<timestamp>

The Convex backend verifies the signature on-chain against ownership of the skill receipt before serving the code bundle. Everything is verified, localized, and secure.