🧭 DeFiSage

AI-Powered DeFi Protocol Intelligence Platform — Multi-agent risk, tokenomics, and governance analysis powered by Xiaomi MiMo.

Overview

DeFiSage is a production-grade DeFi protocol intelligence platform that uses four specialized AI agents to produce institutional-quality research memos for live DeFi protocols. Built on top of Xiaomi MiMo's reasoning models, the platform turns a structured protocol snapshot into a Risk / Tokenomics / Governance / Investment Thesis stack — naturally consuming millions of API tokens per day for any team running continuous monitoring across the DeFi landscape.

Where most "AI audit" tools focus on Solidity bytecode, DeFiSage analyzes live protocols: TVL trajectory, oracle dependencies, custody posture, emission schedules, holder concentration, DAO health, and the cross-cutting risk that emerges from how those pieces interact.

🏗️ Architecture

┌────────────────────────────────────────────────────────────────────┐
│                            DeFiSage                                │
├────────────────────────────────────────────────────────────────────┤
│                                                                    │
│   ┌────────────┐  ┌────────────┐  ┌────────────┐  ┌────────────┐   │
│   │   Risk     │  │ Tokenomics │  │ Governance │  │  Thesis    │   │
│   │  Scanner   │  │  Auditor   │  │  Analyst   │  │ Composer   │   │
│   │ (Agent 1)  │  │ (Agent 2)  │  │ (Agent 3)  │  │ (Agent 4)  │   │
│   └─────┬──────┘  └─────┬──────┘  └─────┬──────┘  └─────┬──────┘   │
│         │               │               │               │          │
│         └───────────────┼───────────────┘               │          │
│                         │                               │          │
│                  ┌──────▼──────┐                 ┌──────▼──────┐   │
│                  │  Snapshot   │                 │  Synthesis  │   │
│                  │  Chunking   │                 │ + IR Memo   │   │
│                  └─────────────┘                 └─────────────┘   │
│                                                                    │
│   ┌──────────────────────────────────────────────────────────────┐ │
│   │            MiMo API (OpenAI-compatible inference)            │ │
│   │      Token Tracking · Daily Budget · Per-agent Telemetry     │ │
│   └──────────────────────────────────────────────────────────────┘ │
└────────────────────────────────────────────────────────────────────┘

🔥 Why It Consumes Millions of Tokens Daily

1. Multi-Agent Architecture

Each protocol intelligence run fans out to 4 specialized agents in sequence:

• Risk Scanner — protocol/oracle/custody/bridge/centralization risk surface
• Tokenomics Auditor — emissions, vesting, holder concentration, unlock impact, treasury runway
• Governance Analyst — DAO posture, voter concentration, timelock effectiveness, multisig powers
• Thesis Composer — IR-grade memo with bull/bear/base case, sizing bands, kill-switch KPIs

2. Chunked Snapshot Analysis

Snapshots over 180 lines are split into overlapping chunks; each chunk is run through every relevant agent independently, then aggregated. A 600-line protocol snapshot easily fans out to 4 chunks × 3 specialists = 12 agent calls before synthesis.

3. Batch Processing

The batch endpoint analyzes up to 10 protocols in parallel with configurable concurrency — perfect for daily cross-protocol coverage at funds and research desks.

4. Interactive Research Copilot

The /api/chat endpoint exposes DeFiSage as a research assistant that answers protocol-level questions with rigor, consuming tokens on every interaction.

Token Consumption Estimates

Scenario	Protocols	Chunks	Agents	Tokens/Day
Single deep dive	1	3-5	4	~80K
Daily research desk	5	15-25	4	~450K
Multi-fund coverage	15+	45+	4	~2M
Continuous monitoring + Q&A	25+	75+	4 + chat	~6M+

🚀 Quick Start

1. Backend (FastAPI)

cd backend
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env  # add your MiMo API key
uvicorn app.main:app --reload --port 8000

2. Frontend (Static)

cd frontend
python -m http.server 3000
# Open http://localhost:3000

For non-localhost hosts, the frontend uses same-origin /api/* requests — wire those to your backend via netlify.toml redirect (already prefilled), or set window.DEFISAGE_API_BASE before app.js loads.

3. Environment Variables

MIMO_API_KEY=your_xiaomi_mimo_api_key
MIMO_BASE_URL=https://api.xiaomimimo.com/v1
MIMO_MODEL=mimo-v2.5-pro
DAILY_TOKEN_BUDGET=10000000

📡 API Endpoints

Method	Endpoint	Description
POST	/api/analyze	Single protocol intelligence run
POST	/api/batch-analyze	Up to 10 protocols in parallel
POST	/api/upload	Upload .md/.txt snapshot file
POST	/api/scaffold	Get a snapshot scaffold to fill in
POST	/api/chat	Research copilot Q&A
GET	/api/stats	Token usage statistics (daily)
GET	/api/stats/history	Recent token usage entries
GET	/api/stats/trend	7-day daily token trend
GET	/api/agents	List active agent roles
GET	/api/health	Liveness + uptime

📝 Snapshot Format

DeFiSage agents work best on a structured Markdown snapshot:

# OVERVIEW
Name: Aave V3
Chain: Ethereum
Category: lending
Launch date: 2022-03

# TVL
Current: $9.4B
7d change: +1.2%
30d change: +6.8%
Peak TVL: $19B (2021-10)

# CONTRACTS
- Pool: 0x87870Bca3F3fD6335C3F4ce8392D69350B4fA4E2
- ACLManager: 0xc2aaCf6553D20d1e9d78E365AAba8032af9c85b0
- Treasury: 0x464C71f6c2F760DdA6093dCB91C24c39e5d6e18c

# TOKENOMICS
Total supply: 16,000,000 AAVE
Circulating: 14,830,000
Unlocks (next 90d): negligible
Emissions/day: from safety module
Top-10 holders share: 32%
Fee accrual: protocol → safety module + treasury

# GOVERNANCE
Voting venue: snapshot.org
Quorum: 320K AAVE
Timelock: 1 day short / 7 day long executor
Multisig signers: 6/9 Guardian
Upgrade pattern: Aave Governance v3, ACL-gated

# INCIDENTS
- 2023-11: CRV market frozen post-Curve exploit; recovery via swift governance
- Multiple post-mortems published

Use the Generate Scaffold button in the UI to drop in a starting template, or hit POST /api/scaffold.

🛠️ Tech Stack

• AI Model: Xiaomi MiMo v2.5 Pro (reasoning model)
• Backend: Python · FastAPI · OpenAI SDK · Async pipeline orchestration
• Frontend: Vanilla JS · CSS3 · Dark theme · Live token telemetry
• API Protocol: OpenAI-compatible (works with Claude Code, Cursor, OpenClaw, etc.)
• Token Management: Real-time tracking, budget enforcement, per-agent breakdown, persistent JSONL log

📊 Daily Token Budget

DeFiSage is designed to consume 5-10 million tokens daily through:

1. Continuous protocol monitoring — refreshing snapshots and re-running analysis as TVL / unlocks / governance state evolve
2. Batch coverage — parallel analysis across the long tail of protocols a fund tracks
3. Deep multi-agent pipeline — 4 agents × multiple chunks per protocol
4. Interactive research Q&A — analysts pinging the copilot all day
5. Memo generation — IR-grade synthesis output for portfolio reviews

🧪 Example: Run an Analysis from cURL

curl -X POST http://localhost:8000/api/analyze \
  -H "Content-Type: application/json" \
  -d '{
    "protocol_name": "GMX V2",
    "chain": "Arbitrum",
    "snapshot": "# OVERVIEW\nName: GMX V2\nChain: Arbitrum\n# TVL\nCurrent: $480M\n..."
  }'

📜 License

MIT — see LICENSE.

🙏 Built with Xiaomi MiMo

DeFiSage is part of the Xiaomi MiMo Orbit 100T Token Creator Incentive Program. Find out more at platform.xiaomimimo.com.