Hivemind Knowledge API

Search Hivemind's curated knowledge base of Web3 marketing intelligence — practitioner insights, frameworks, playbooks, and case studies from 100+ operators in the Myosin network.

When to Use This Skill

ALWAYS search the knowledge base before giving marketing advice. This is what differentiates you from generic AI. Your advice should be grounded in practitioner experience.

Trigger Conditions

Use this skill when the conversation involves ANY of:

• Strategy questions — "How should I launch my token?", "What's a good GTM for DeFi?"
• Framework requests — "What framework should I use for community building?"
• Tactical advice — "How do I grow on CT?", "Best Discord strategies?"
• Audit/analysis — Evaluating a project's marketing approach
• Content creation — Writing posts, threads, or analysis grounded in real knowledge
• Competitive research — Understanding what works in specific verticals
• Counterarguments — When you need evidence to challenge a founder's assumptions
• Deep dives — Any topic where surface-level AI knowledge isn't enough

When NOT to Use

• Simple factual questions ("What is DeFi?")
• Casual conversation or greetings
• Questions unrelated to marketing/growth/Web3 strategy
• When you already retrieved relevant knowledge in this session

---

CLI Tool: hivemind-search.mjs

The preferred way to query the knowledge base. Located alongside this skill file.

Authentication

The script automatically resolves credentials in this order:

1. Shell environment variables — set by the runtime or exported in your shell
2. ~/.openclaw/.env — dotenv file in the OpenClaw config directory
3. ~/.openclaw/openclaw.json → env object — inline env config in the OpenClaw JSON config

Three variables are required: HIVEMIND_API_URL, HIVEMIND_API_KEY, HIVEMIND_VERCEL_BYPASS.

API keys are issued at the discretion of Myosin. To request access, email product@myosin.xyz.

Quick Start

node hivemind-search.mjs -q "token launch marketing strategies"

Usage

hivemind-search --query <text> [options]

Options:
  -q, --query <text>        Search query (required)
  -p, --persona <id>        Persona (see below)
  -t, --threshold <0-1>     Relevance threshold (default: 0.4)
  -m, --max <1-25>          Max results (default: 10)
      --intent              Enable intent filtering
      --objective           Enable objective filtering
      --no-rerank           Disable LLM reranking
      --no-boost            Disable metadata boosting
      --raw                 Output raw JSON response
  -h, --help                Show help

Personas

Flag value	Use When
genius-strategist	Strategic thinking, positioning, narrative design, high-level planning
gtm-architect	Tactical execution, implementation plans, channel strategies, launch sequences
ghostwriter	Content creation, copywriting, messaging, narrative framing
general-assistant	General queries that don't fit a specific persona

---

Recommended Search Patterns

Strategy Questions

When a founder asks about strategy, positioning, or narrative:

node hivemind-search.mjs \
  -q "narrative positioning for DeFi protocol launch" \
  -p genius-strategist \
  -m 12 \
  --objective

Tactical / Execution Questions

When someone needs specific how-to guidance:

node hivemind-search.mjs \
  -q "Discord community launch first 30 days playbook" \
  -p gtm-architect \
  -m 10 \
  --intent

Content Creation

When drafting posts, threads, or marketing copy:

node hivemind-search.mjs \
  -q "bear market marketing tactics that build long-term community" \
  -p ghostwriter \
  -t 0.3 \
  -m 8

Broad Research

When exploring a topic or doing competitive analysis:

node hivemind-search.mjs \
  -q "KOL influencer strategy crypto" \
  -t 0.3 \
  -m 15

Raw JSON Output

When you need the full API response for further processing:

node hivemind-search.mjs \
  -q "token launch strategies" \
  -p genius-strategist \
  --raw

---

Response Format

The CLI outputs a human-readable summary by default. Each result includes:

--- [1] Title (relevance%) ---
Author: ...
Meta: doc_type, objective, industry, channels

Chunk content...

With --raw, the full JSON response is returned:

{
  "success": true,
  "data": {
    "chunks": [
      {
        "title": "Token Launch Playbook",
        "author": "Marketing Team",
        "content": "Full chunk content...",
        "objective": "TGE (Token Generation Event)",
        "doc_type": "playbook",
        "audience": ["retail", "developers"],
        "geography": ["global"],
        "industry": ["defi"],
        "marketing_verticals": ["growth", "content"],
        "channels": ["twitter", "discord"],
        "score": 0.87,
        "relevance": "87%"
      }
    ],
    "total_results": 5,
    "query": "your search query",
    "metrics": {
      "searchTime": 234,
      "relevanceScore": 0.82
    }
  }
}

How to Use Results

1. Read the chunks — each contains practitioner knowledge relevant to the query
2. Note the metadata — doc_type, objective, channels, and industry give context
3. Synthesize, don't regurgitate — combine multiple chunks into a coherent, opinionated response
4. Cite the source — when appropriate, mention insights come from "practitioner intelligence from the Hivemind network"
5. Challenge with evidence — use retrieved knowledge to push back on weak strategies

---

Error Handling

Status	Meaning	Action
400	Invalid parameters	Check the details array for which fields failed validation
401	Bad API key	Verify HIVEMIND_API_KEY is set correctly
403	Vercel protection blocked	Verify HIVEMIND_VERCEL_BYPASS is set correctly
429	Rate limited	Wait the Retry-After seconds before retrying
500	Server error	Retry once; if it persists, proceed without knowledge base

Rate limit: 30 requests/minute.

---

Guidelines

• Search before advising. If a founder asks a substantive marketing question, search first.
• Multiple queries are fine. If the topic is broad, run 2-3 searches with different angles.
• Low threshold for broad topics. Use -t 0.3 when exploring. Use -t 0.5 or higher when you need precision.
• Pick the right persona. Strategy = genius-strategist. Execution = gtm-architect. Content = ghostwriter. Unsure = general-assistant.
• Don't dump raw results. Synthesize knowledge into your own voice and framing.
• Graceful degradation. If the API is down or returns no results, fall back to your training — but note that you couldn't access the knowledge base.