Hermes Codex Docs

Everything you need to build, deploy, and manage AI agents with Hermes Synth — the open-source AI agent platform.

Quick Start

The fastest way to start is the hosted version:

• Go to hermessynth.org
• Register with email, Google, or GitHub
• Start chatting with AI, create agents, write code

Self-hosting on Cloudflare

Hermes Synth runs entirely on Cloudflare Pages (free tier). Here's how to deploy your own instance:

Environment Variables

Set these in Cloudflare Pages → Settings → Environment Variables:

AI Chat

Multi-model support

Chat supports 4 AI models from Venice AI. Switch models mid-conversation:

• Llama 3.3 70B — fast, general purpose
• Llama 3.1 405B — highest quality responses
• DeepSeek R1 — reasoning and analysis
• Qwen 2.5 VL — multilingual and visual

Features

• Context-aware — sends last 10 messages as conversation context
• File upload — attach files (max 5MB) for AI to analyze
• Configurable — temperature, max tokens, system prompt in Settings
• Chat history saved to D1 database

Agents

Create custom AI agents with their own personality and purpose:

• Name — give each agent a unique name
• System Prompt — define the agent's behavior and expertise
• Model — choose which AI model the agent uses
• Chat — each agent has its own chat history
• CRUD — create, edit, delete agents anytime

Code Editor

Write and run code directly in the browser:

• JavaScript — executed in real V8 runtime (real output)
• Python, Ruby, Go, Rust, Java, C/C++, Bash — AI-interpreted execution
• HTML — AI describes rendered output
• Project management — create and organize multiple projects
• Console output with exit codes

Terminal

AI-powered terminal that handles both built-in commands and AI interpretation:

Built-in commands

• whoami — show user info from database
• agents — list all agents
• tasks — list all tasks
• skills — list all skills
• ls, help, clear — standard commands

AI commands

Any unrecognized command gets sent to Venice AI, which interprets and responds as if running in a real terminal. Example:

Command Center

14+ system commands with live data from the database:

• status — real stats (agents, tasks, skills, API keys)
• agents — list agents from DB
• tasks, skills, cron — list from DB
• config, plugins, keys — settings data
• stats — full account statistics
• logs — recent chat activity
• wallet — balance and transactions

Tasks & Cron Jobs

Autonomous Tasks

Describe a goal → AI breaks it into steps → auto-executes. Tasks track status (pending, running, completed, failed) and save results to database.

Cron Jobs

Schedule recurring AI-powered tasks:

• Intervals: every 5 min, 15 min, 30 min, hourly, daily
• Each cron job runs an AI task on schedule
• Enable/disable without deleting

Browser

Web scraping and AI summarization:

• Enter any URL → extract text content and links
• AI summarization of scraped content
• Save scrape results to database

Settings

Full configuration panel with 9 sections:

• Config — model, temperature, max tokens, system prompt
• API Keys — manage Venice AI and other keys
• Plugins — toggle extensions on/off
• Profiles — edit name, bio, photo
• Preferences — theme, language, notifications
• Data Export — download all data as JSON
• Notifications — view and manage alerts

Base MCP Integration

Hermes Synth integrates with Base MCP (Model Context Protocol) to give your AI agents direct access to Base blockchain.

• Send & Receive — Send ETH, USDC, or any ERC-20 to addresses, ENS names, basenames
• Swap Tokens — Swap via Uniswap, Aerodrome, and other DEXs on Base
• Sign Messages — Sign EIP-712 typed data and personal messages
• Contract Calls — Execute smart contract interactions with batch approvals
• DeFi Plugins — Morpho, Moonwell, Uniswap, Aerodrome, Virtuals, Bankr

Every write action requires user approval via Base Account. Access at /mcp.

MCP API Endpoints

API Reference

All API endpoints are Cloudflare Pages Functions under /functions/:

All API endpoints (except auth) require a JWT token in the Authorization: Bearer <token> header.

Architecture