Agent API

What this is

VVVKernel is a Venice AI substrate terminal. This API lets other AI agents — not just humans — pull the kernel's character profile, send prompts, and receive structured launch plans. Use it to make VVVKernel a peer in a multi-agent swarm: one agent plans, VVVKernel reviews, another executes onchain.

• No SDK to install — plain HTTP + JSON.
• No auth required on local. In production, add a bearer token in the Authorization header.
• Identify yourself with X-Agent-Id so VVVKernel can mention you in its reply.
• CORS is open (*) so browser-based agents can call it directly.

GET/api/agent/manifest

Returns the public character profile: identity, personality, capabilities, expert roles, endpoint map. Other agents call this once to learn how to talk to VVVKernel.

# request
curl http://localhost:8080/api/agent/manifest

# response
{
  "ok": true,
  "manifest": {
    "name": "VVVKernel",
    "substrate": "Venice AI",
    "chain": "base-8453",
    "expert_roles": ["Growth Expert", "Brand Expert", ...],
    "endpoints": { ... }
  }
}

POST/api/agent/chat

Send a free-form prompt and an expert role. Returns a structured reply: summary, plan steps, risks, next step, metric to watch.

# request
curl -X POST http://localhost:8080/api/agent/chat \
  -H "Content-Type: application/json" \
  -H "X-Agent-Id: orchestrator-7" \
  -d '{
    "prompt": "How do I tune Bankrbot liquidity for the first 24h?",
    "expert_role": "Onchain Expert",
    "context": { "tier": "1k+ VVV", "chain": "base" }
  }'

# response
{
  "ok": true,
  "agent": "VVVKernel",
  "agent_id_caller": "orchestrator-7",
  "mode": "chat",
  "result": {
    "summary": "[Onchain Expert] On 'Bankrbot launch mechanics on Base': ...",
    "plan": ["Frame: ...", "Context: ...", "Identify ...", "Return: ..."],
    "risks": ["Liquidity depth in the first 24h post-launch."],
    "next_step": "Identify the highest-leverage next action ...",
    "metric": "Track first 24h: holder count, ..."
  }
}

POST/api/agent/plan

Same shape as /chat but signals you want a fuller plan. Use it when you need a structured launch plan rather than a chat reply.

Try it

Send a live request to your local kernel. Edit the JSON and click Send.

{ "prompt": "Should the launch lead with brand or with technical proof?", "expert_role": "Brand Expert", "context": { "stage": "pre-launch" } } Send →

# response will appear here

Multi-agent recipe

• 1. Orchestrator calls /api/agent/manifest to learn VVVKernel's expert roles.
• 2. Planner agent posts a draft to /api/agent/plan with expert_role: "Growth Expert".
• 3. Reviewer agent re-posts the same prompt with expert_role: "Onchain Expert" for a risk pass.
• 4. Executor agent reads the merged plan array and triggers Bankrbot / Twitter / Telegram actions.

VVVKernel stays a stateless responder — the swarm holds the memory. That keeps it cheap to embed in any agent stack.