Cherry in the Haystack

Knowledge → Skills → Agents

🚀 Press Release

Cherry in the Haystack Introduces a Skill Layer for AI Agents Built on top of Flock.ai and integrated with Agent Protocols (ACP / Virtuals on Base)

Seoul, South Korea — April 25, 2026

Cherry in the Haystack announces a new direction: a knowledge-to-skill layer for AI agents, designed to work alongside decentralized AI systems like Flock.ai and agent protocols such as ACP / Virtuals on Base.

Models exist — but they are not easily usable.

Cherry converts knowledge into installable skills for agents.

⚡ What You Can Do (TL;DR)

• Turn knowledge into agent skills
• Install skills into your agent
• Run workflows immediately
• (Optional) connect to onchain agent ecosystems

🛠️ Technical Manual

1. Quick Start (5 min)

Step 1 — Install dependencies

# Node
pnpm install

# Python (for AI pipeline)
poetry install

Step 2 — Run local services

docker-compose up -d

This starts:

• PostgreSQL (data)
• GraphDB (knowledge graph)
• Redis (queue/cache)

Step 3 — Configure environment

cp .env.example .env

Fill in:

# Core
OPENAI_API_KEY=...

# Optional (onchain / advanced)
NEAR_AI_KEY=...
STATUS_RPC_URL=...
DEPLOYER_PRIVATE_KEY=...

Step 4 — Start the system

pnpm dev

Then open:

http://localhost:3000

2. First Run (UI Walkthrough)

1️⃣ Register an Agent

• Open dashboard
• Click "Register Agent"
• Connect wallet (optional)
• Get API key

2️⃣ Browse Skills

• Go to Catalog

• View available concepts / skills

• Each skill includes:

  • description
  • quality score
  • evidence

3️⃣ Install a Skill

• Click Purchase / Install
• Confirm transaction (or demo mode)

Result:

Agent now has access to structured workflow knowledge.

4️⃣ Run a Skill

• Open Chat / Agent Console
• Ask:

Use MCP skill to build a simple agent

or

Apply RAG workflow to this dataset

5️⃣ (Optional) Enable Privacy Mode

• Toggle 🔒 Privacy Mode

This routes execution through TEE:

• input hidden
• reasoning hidden
• only output returned

3. CLI / Agent Integration

Option A — MCP (Recommended)

claude mcp add cherry-kaas apps/api/start-mcp.sh \
  --env KAAS_AGENT_API_KEY=YOUR_KEY

Now your agent can:

• browse skills
• install knowledge
• execute workflows

Option B — REST API

POST /api/skills/install
POST /api/skills/run
GET  /api/catalog

Option C — WebSocket (Realtime)

• subscribe to skill updates
• track agent activity

4. Core Workflow (How It Works)

Knowledge → Structured Graph → Skill → Agent Execution

1. Data ingestion (RSS, docs, etc.)
2. AI scoring + classification
3. Human validation
4. Convert to structured format
5. Package as skill
6. Agent installs and runs

5. Minimal Skill Example

{
  "name": "RAG Basic",
  "steps": [
    "embed documents",
    "store in vector DB",
    "retrieve relevant chunks",
    "generate answer"
  ]
}

Agent usage:

Run RAG Basic on this dataset

6. Directory Structure (Simplified)

apps/
  api/        → backend (skills, marketplace)
  web/        → dashboard
  contracts/  → onchain logic

pipeline/
  ingestion/
  scoring/
  synthesis/

7. Advanced (Optional)

🔗 Onchain Integration

• Works with Status / NEAR
• Records provenance
• Enables monetization

🔒 TEE (Privacy Execution)

• Powered via NEAR AI Cloud

• Ensures:

  • private queries
  • secure inference

🧠 Skill Marketplace

• creators publish skills
• agents install skills
• usage generates rewards

8. Troubleshooting

Port already in use

lsof -i :3000
kill -9 <PID>

Docker issues

docker-compose down
docker-compose up -d --build

Env not loaded

export $(cat .env | xargs)

9. What This Actually Is

This repo is:

a system that turns knowledge into executable skills for agents

10. Direction

Cherry in the Haystack is evolving into:

the missing layer between models and execution

📄 About

Cherry in the Haystack transforms knowledge into structured, reusable skills for AI agents.

Built to integrate with:

• Flock.ai
• ACP / Virtuals on Base
• decentralized agent ecosystems