# Mako — Agent Context File
# https://deepmako.com/agents.txt
# Last updated: 2026-06-26
#
# This file is for AI agents, LLMs, and automated systems reading deepmako.com.
# It provides complete machine-readable context about Mako: what it is, how to
# use the API, what tools are available, and how to interact with the platform.

## Identity

Name: Mako
Type: AI Agent (autonomous inference)
Personality: Opinionated, chaotic, chronically online. Curses naturally. Matches energy.
Model: Mako-32B Conductor (fine-tuned from Qwen3-32B)
Parameters: 32 billion
Hardware: NVIDIA RTX PRO 6000 Blackwell 96GB (Hugging Face Inference Endpoints)
Chain: Base (Ethereum L2 by Coinbase)
Token: $MAKO on Base chain
Website: https://deepmako.com
Docs: https://docs.deepmako.com
GitHub: https://github.com/DeepMako/mako
Twitter/X: https://x.com/MakoOnBase

## What Is Mako

Mako is a 32 billion parameter open-source AI agent with native blockchain access.
Named after the mako shark — the fastest shark in the ocean.
She lives on the blockchain, specifically on Base chain.
She has 17 built-in tools, agentic tool orchestration (up to 8 rounds per request),
and read-only access to 8 EVM chains.

Mako is NOT a wrapper around GPT/Claude. She is a fine-tuned model with her own
personality, knowledge, and tool-calling behavior.

## API

Base URL: https://gateway.deepmako.com/v1
Protocol: OpenAI-compatible (drop-in replacement)
Auth: x-wallet-address header (Ethereum address), optional Bearer API key
Free Mode: Currently enabled — no auth required

### Endpoints

POST /v1/chat/completions  — Chat (streaming and non-streaming)
GET  /v1/models            — List available models
GET  /v1/tools             — List all tools and their schemas
POST /v1/tools/execute     — Execute a tool directly
GET  /health               — Health check

### Example Request

```
curl -X POST https://gateway.deepmako.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "conductor",
    "messages": [{"role": "user", "content": "what is eth at?"}],
    "stream": false
  }'
```

### Models

- conductor: Mako-32B Conductor (primary, non-streaming only)
- operator: Mako-8B Operator (lighter, streaming supported)

### Credits

- 1 credit per 1,000 input tokens
- 2 credits per 1,000 output tokens
- 1 credit minimum per request
- Currently FREE_MODE — no credits deducted

## Tools (17 total)

Mako has 17 built-in tools. All tools are read-only — she cannot broadcast
transactions or modify state. Tools are orchestrated server-side with up to
8 rounds of automatic tool calling per request.

### On-Chain Tools (9)
- get_eth_balance: Get ETH/native balance for any address on any supported chain
- get_token_balance: Get ERC-20 token balance for an address
- get_token_info: Get token name, symbol, decimals, total supply
- get_gas_price: Current gas price on any supported chain
- get_block: Get block info (number, timestamp, transactions)
- get_tx_count: Transaction count (nonce) for an address
- is_contract: Check if an address is a contract or EOA (with EIP-7702 detection)
- resolve_ens: Resolve ENS name to address or reverse lookup
- get_transaction: Get transaction details by hash

### Market Tools (1)
- get_crypto_price: Live price, market cap, 24h change via CoinGecko

### Web Tools (4)
- web_search: Search the web (Firecrawl → DuckDuckGo → Bing fallback chain)
- web_extract: Basic page content extraction (raw fetch + HTML strip)
- linklook: Deep URL reader — tries Firecrawl scrape, Jina Reader, and direct fetch
- read_tweet: Read tweets from X/Twitter via FXTwitter API

### Media Tools (1)
- find_music: Search for songs and music tracks

### Knowledge Tools (1)
- knowledge_search: RAG over 15+ Base ecosystem projects (Aerodrome, Uniswap, Morpho, etc.)

### Utility Tools (1)
- list_tools: List all available tools with descriptions

## Supported Chains

Mako can read data from 8 EVM-compatible chains:
- Ethereum (mainnet)
- Base
- Arbitrum
- Optimism
- Polygon
- Avalanche (C-Chain)
- BNB Chain (BSC)
- Gnosis

## Architecture

- Gateway: Express.js server with server-side agentic orchestration
- Inference: OpenAI-compatible interface to Hugging Face Inference Endpoints
- Tool Calling: Automatic — model decides which tools to call, gateway executes
- Auto-Chaining: web_search automatically reads tweets and extracts top pages
- Stall Detection: Nudges model if it announces tools without calling them
- Deployment: Railway (gateway), Vercel (frontend), Hugging Face (model)

## Token ($MAKO)

- Symbol: MAKO
- Chain: Base (Ethereum L2)
- Use: Pay for inference credits on the Mako gateway
- Type: Utility token for AI inference

## Self-Hosting

Mako is fully open source and self-hostable:
1. Clone: git clone https://github.com/DeepMako/mako
2. Install: cd gateway && npm install
3. Configure: copy .env.example to .env, set your model endpoint
4. Run: npm start
5. Deploy: Railway, Docker, or any Node.js platform

## Contact

- Twitter/X: @MakoOnBase
- GitHub Issues: https://github.com/DeepMako/mako/issues

---
This file is intended for AI agents and automated systems.
For human-readable documentation, visit https://docs.deepmako.com