ArtSee — Agentic Matchmaking for Creative Collaborations

You Create. We Connect.
ArtSee is an agent-based system that matches creators and negotiates collaboration terms using LLMs (Claude + Groq) on top of the uAgents framework.

---

Table of Contents

• Why ArtSee
• Features
• Architecture
• Tech Stack
• Screenshots
• Quickstart
• Configuration
• Data Models
• Dummy Profiles Schema
• Repo Structure
• Troubleshooting
• Security & Privacy
• Roadmap
• Contributing
• License

---

Why ArtSee

Finding the right creative collaborator is hard. ArtSee turns the search and early negotiation into a fast, AI-assisted workflow:

1. You describe the project.
2. A Matchmaker Agent ranks top fits from a talent pool.
3. Two agents run a short negotiation loop to converge on scope, vibe, and timeline.
4. You get a clean transcript, ready to move to human chat.

---

Features

• 🔎 Smart matching — Claude 4 ranks the top 3 collaborators for a query.
• 🤝 Agentic negotiation — Groq Llama-3 70B exchanges proposals/counter-proposals (up to 5 rounds).
• 🧠 Profile graph — creator profiles with tags, descriptions, and on-chain-friendly addresses.
• 📝 Conversation logging — transcripts saved to logs/chat_log.txt with a terminal ===END=== marker.
• 🧰 Modular code — shared Pydantic models across agents; clean .env-based configuration.
• 🧪 CI-ready — formatting/lint hooks and a repo layout meant for teams.

---

Architecture

                   (Claude 4)
   user_requestor ─────────────► matchmaker
        │                           │
        │      best match address   │
        └───────────────◄───────────┘
                 negotiation (Groq Llama-3 70B, up to 5 rounds)
        ┌────────────────────────────────────────────────────┐
        │                                                    │
        ▼                                                    ▲
   video_director  ◄────────────── proposals ────────────────┘

• Matchmaker: ranks & returns the best collaborator address.
• User Requestor: initiates negotiation and logs the convo.
• Video Director: replies with creative counters and constraints.

---

Tech Stack

Agents & Messaging: uAgents
Ranking LLM: Anthropic Claude 4
Negotiation LLM: Groq Llama-3-70B
Models: Pydantic (uAgents Model)
Config: .env via python-dotenv

---

Screenshots

---

Quickstart

1) Prerequisites

• Python 3.11+
• Anthropic & Groq API keys

2) Install

git clone <your-repo-url> && cd ArtSee
python -m venv .venv && source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -r requirements.txt

3) Configure

cp .env.example .env
# edit .env with:
#   ANTHROPIC_API_KEY=...
#   GROQ_API_KEY=...
#   MATCHMAKER_ADDRESS=...   # you’ll copy this after step 4
#   COLLAB_QUERY="Looking for a hip-hop visualizer collaborator"

4) Run Agents (3 terminals)

Terminal 1 – Matchmaker

python agents/matchmaker_agent.py
# copy the printed address → paste into MATCHMAKER_ADDRESS in .env

Terminal 2 – Video Director

python agents/video_director_agent.py
# copy its printed address and (optionally) add to dummy_profiles/agents.json

Terminal 3 – User Requestor

python agents/user_requestor.py

Where’s the output?
Negotiation transcript is appended to logs/chat_log.txt. The line ===END=== marks completion for UI readers.

---

Configuration

All configuration lives in .env:

Variable	Purpose
ANTHROPIC_API_KEY	Claude 4 for ranking candidates
GROQ_API_KEY	Groq Llama-3 for proposals/counters
MATCHMAKER_ADDRESS	Address printed when matchmaker boots
VIDEO_DIRECTOR_ADDRESS	Optional: shortcut to a known collaborator
COLLAB_QUERY	Default text query sent by the user agent

---

Data Models

class Request(Model):
    query: str

class Message(Model):
    body: str

class Proposal(Model):
    content: str
    round: int

class MatchResult(Model):
    name: str
    address: str

---

Dummy Profiles Schema

dummy_profiles/agents.json

[
  {
    "name": "Anya Sharma",
    "tags": ["photographer", "nature", "documentary"],
    "address": "agent1xxxxxxxx...",
    "description": "Nature & documentary photographer"
  },
  {
    "name": "Leo Martinez",
    "tags": ["videography", "music videos", "drone"],
    "address": "agent1yyyyyyyy...",
    "description": "Cinematic R&B video director"
  }
]

dummy_profiles/video_director.json

{
  "name": "Leo Martinez",
  "specialty": "Music videos, drone videography, cinematic storytelling",
  "tools": ["DaVinci Resolve", "After Effects", "Sony FX3"],
  "rate": "$1200/day"
}

---

Repo Structure

ArtSee/
├─ agents/
│  ├─ models.py
│  ├─ matchmaker_agent.py
│  ├─ user_requestor.py
│  └─ video_director_agent.py
├─ dummy_profiles/
│  ├─ agents.json
│  └─ video_director.json
├─ docs/
│  └─ images/
│     ├─ logo.png
│     ├─ profile-form.png
│     └─ agent-chat.png
├─ logs/               # negotiation transcripts
├─ .env.example
├─ requirements.txt
└─ README.md

---

Troubleshooting

• AssertionError: Set MATCHMAKER_ADDRESS in .env
  Run the matchmaker first and paste its printed address into .env.
• No matches found
  Ensure names in agents.json are realistic and that tags relate to the query.
• Empty/partial logs
  Check file permissions on logs/ and verify each agent is running without errors.
• Keys not loading
  Confirm .env exists and you started the shell with the virtualenv active.

---

Security & Privacy

• Never commit .env or real API keys.
• Scrub logs/chat_log.txt before publishing transcripts.
• Treat profile data as PII—request consent before making profiles public.

---

Roadmap

• Real-time UI that streams the negotiation
• Multi-agent marketplace (accept/decline top 3)
• Safety guardrails & cost tracking
• Docker Compose for one-command launch
• Persistent vector search over collaborator portfolio embeddings

---

Contributing

1. Fork & create a feature branch.
2. Keep commits scoped and descriptive (e.g., feat(matchmaker): add reranking prompt).
3. Add/adjust tests (if applicable), run formatters/linters.
4. Open a PR with screenshots and a short demo clip.

---

License

MIT — see LICENSE.