VVVKernel Docs

Kernel modes

Three execution modes share the same substrate but differ in output discipline. Pick the mode that matches your task.

The kernel never blends modes. A plan request always returns a numbered plan; a chat request never gets cut to a list unless asked.

Holder tiers

Tiers are pre-launch — final thresholds are set on TGE. The kernel advises against promising any unverified perk.

CLI · Install

Published to npm as vvvkernel-cli. Requires Node 18+.

npm install -g vvvkernel-cli
vvvkernel --version
# 1.0.1

Or one-shot via npx:

npx vvvkernel-cli ask "what is the kernel?"

CLI · Commands

All commands respect --json for machine-readable output.

CLI · Custom skills

Drop a markdown file in ~/.vvvkernel/skills/. The CLI loads it as a callable skill on the next run.

# ~/.vvvkernel/skills/audit-checklist.md
---
name: audit-checklist
description: Run a security checklist on a contract address
---

You are an audit lead. Given a contract address:
1. Check ownership and proxy status
2. Verify source on Etherscan
3. List all privileged functions
4. Flag honeypot signals
5. Output a one-page report

Skills are local-first. Anything you add stays on your machine unless you opt in to publishing.

CLI · Config

Configuration is read from ~/.vvvkernel/config.json with env overrides taking precedence.

{
  "endpoint": "https://vvvkernel.com",
  "model": "venice-substrate",
  "stream": true,
  "timeout_ms": 60000
}

Env vars: VVVKERNEL_ENDPOINT, VVVKERNEL_MODEL, VVVKERNEL_TOKEN.

MCP · Overview

VVVKernel speaks the Model Context Protocol. Any MCP-aware client (Claude Desktop, Cursor, Continue, custom) can mount the kernel as a tool surface.

Two transports are supported:

• stdio — local CLI process spawned by the host. Lowest latency, zero network setup.
• http — remote JSON-RPC over HTTP. Use this for shared servers, multi-seat teams, or when the host doesn't support stdio.

The protocol exposes three core methods: initialize, tools/list, and tools/call. The kernel registers tools for chat, plan, expert routing, and skill execution.

MCP · Claude Desktop

Edit claude_desktop_config.json:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "vvvkernel": {
      "command": "npx",
      "args": ["-y", "vvvkernel-cli", "mcp"],
      "env": {
        "VVVKERNEL_ENDPOINT": "https://vvvkernel.com"
      }
    }
  }
}

Restart Claude Desktop. The kernel appears in the tool list. Try: "Ask vvvkernel to draft a launch plan for a tier-1 release."

MCP · Cursor

Open Cursor → Settings → MCP → Add new server. Use the same shape as Claude Desktop:

{
  "vvvkernel": {
    "command": "npx",
    "args": ["-y", "vvvkernel-cli", "mcp"]
  }
}

Tools become callable from the agent panel. Useful inside Cursor: plan for refactor sequencing, experts to route a code question to the audit lead.

MCP · HTTP transport

Skip the CLI and connect directly:

curl -X POST https://vvvkernel.com/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{}}}'

List tools:

curl -X POST https://vvvkernel.com/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/list"}'

Call a tool:

curl -X POST https://vvvkernel.com/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"plan","arguments":{"goal":"ship phase-1 audit"}}}'

MCP · Tools reference

API · Health

GET /health
200 OK
{"ok":true,"version":"1.0.1","uptime_s":12345}

API · Chat (SSE stream)

POST /api/chat
Content-Type: application/json

{"messages":[{"role":"user","content":"what is the kernel?"}]}

Response is text/event-stream. Each event is a JSON delta:

data: {"type":"delta","text":"VVVKernel is "}
data: {"type":"delta","text":"a Venice AI substrate..."}
data: {"type":"done"}

API · Agent · chat & plan

Single-turn structured answers. Use these when you don't need streaming.

POST /api/agent/chat
{"input":"route this question to the right expert: tokenomics for a new chain"}

POST /api/agent/plan
{"goal":"launch phase-1 community in 30 days","horizon":"week"}

Both return {"output": string, "meta": {...}}.

API · Manifest

Machine-readable description of the kernel for client discovery:

GET /api/manifest
{
  "name": "vvvkernel",
  "version": "1.0.1",
  "experts": [...],
  "modes": ["chat","agent","plan"],
  "mcp": {"stdio":true,"http":true}
}

Troubleshooting

Still stuck? Open an issue with the request id from the response header (x-vvv-rid).

Changelog

1.0.1 — current

• Fix: vvvkernel experts now formats name + focus instead of [object Object].
• Docs: full /docs page with MCP integration, HTTP API reference, troubleshooting.
• Kernel: enriched system prompts (chat / agent / plan) with substrate context and 7-expert focus.

1.0.0

• Initial public release on npm.
• stdio MCP server, HTTP transport, plan + chat + experts commands.
• Custom skills loader from ~/.vvvkernel/skills/.