Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.apiyi.com/llms.txt

Use this file to discover all available pages before exploring further.

​
Overview

Hermes Agent is an open-source AI agent built by Nous Research, positioned as “the agent that grows with you.” It’s one of the few agents with a built-in learning loop — autonomously creating skills from experience, improving them during use, nudging itself to persist knowledge, searching its own past conversations via FTS5 full-text index, and maintaining a deepening user model across sessions. Hermes doesn’t have to live on your laptop — it runs equally well on a $5 VPS, a GPU cluster, or serverless infrastructure that costs nearly nothing when idle. By integrating APIYI, you get:

🧠 Self-Improving Loop

Autonomous skill creation + refinement, long-term memory across sessions

📱 Multi-Platform Gateway

Telegram / Discord / Slack / WhatsApp / Signal / Email / CLI

⏰ Cron Scheduling

Built-in scheduler delivers reports/audits via any platform

☁️ Runs Anywhere

Seven terminal backends — local / Docker / SSH / Modal / Daytona / Vercel Sandbox
Project Info: Hermes Agent is MIT-licensed open source. Repository: github.com/NousResearch/hermes-agent. Docs: hermes-agent.nousresearch.com/docs/.

​
Install

​
Linux / macOS / WSL2 / Termux

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

​
Windows (PowerShell, early beta for native)

iex (irm https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.ps1)
The installer handles uv, Python 3.11, Node.js, ripgrep, ffmpeg, and a portable Git Bash automatically. Once installed, reload your shell and start: 
source ~/.bashrc    # or source ~/.zshrc
hermes              # Launches the TUI, ready to chat

​
Connect to APIYI

Hermes ships with the hermes model command, supporting Nous Portal / OpenRouter / OpenAI / custom endpoints. APIYI exposes an OpenAI-compatible API — plug it in as a “custom OpenAI endpoint” and you get APIYI’s full model matrix in one go. 
hermes model        # Enter the model selection wizard
The wizard will ask:
StepInput
ProviderPick OpenAI (or Custom OpenAI endpoint)
API Base URLhttps://api.apiyi.com/v1
API KeyYour APIYI key (sk-...)
ModelThe model ID you want, e.g. gpt-5.4, claude-sonnet-4-6, deepseek-v3.2, gemini-3.1-pro-preview

​
Option 2: Configure via hermes config set

hermes config set llm.provider openai
hermes config set llm.base_url https://api.apiyi.com/v1
hermes config set llm.api_key sk-your-apiyi-key
hermes config set llm.model gpt-5.4
Then run hermes once to verify the connection.

​
Option 3: Environment Variables (Best for Docker / Serverless)

export OPENAI_API_BASE=https://api.apiyi.com/v1
export OPENAI_API_KEY=sk-your-apiyi-key
export HERMES_MODEL=gpt-5.4
hermes
Switch models any time via hermes model or by updating HERMES_MODEL — no code changes needed.

​
Option 4: Anthropic Native Protocol (Best for Claude-Heavy Workloads)

Hermes treats Anthropic as a first-class provider. Internally, the wire protocol is called anthropic_messages, and it ships with perks the OpenAI-compatible path doesn’t get:
Why native Anthropic is the better path for Claude: For native Anthropic, OpenRouter, and Nous Portal providers, Hermes automatically attaches cache_control breakpoints with a 1-hour TTL on the system prompt, skill blocks, and the early portion of long context. Subsequent sends across sessions and forked subagents reuse the cache at the discounted cached-read rate. This optimization does NOT kick in on the OpenAI-compatible path.
CLI configuration: 
hermes config set llm.provider anthropic
hermes config set llm.base_url https://api.apiyi.com
hermes config set llm.api_key sk-your-apiyi-key
hermes config set llm.model claude-sonnet-4-6
Environment-variable equivalent: 
export ANTHROPIC_BASE_URL=https://api.apiyi.com
export ANTHROPIC_API_KEY=sk-your-apiyi-key
export HERMES_MODEL=claude-sonnet-4-6
hermes
Do NOT include /v1 in base_url — it must be https://api.apiyi.com. The Anthropic protocol appends /v1/messages automatically; including /v1 yourself produces .../v1/v1/messages and triggers a 404.
Hermes auto-detects the wire protocol from the URL (paths ending in /anthropic route to anthropic_messages). For non-standard endpoints like LiteLLM proxies, set the mode explicitly: 
hermes config set llm.api_mode anthropic_messages
Bonus: When creating a token at api.apiyi.com/token, pick the ClaudeCode group to automatically get a 5% discount, stackable with the 10%-20% top-up bonus.
Why APIYI?
  • One key, many models: OpenAI / Anthropic / Google / DeepSeek / Zhipu / Kimi and more
  • Pricing advantage: typically 5%-20% off vs. official pricing, with deposit bonuses on selected models
  • Direct access in China: reach overseas LLMs without a VPN
  • Dual-protocol compatibility: both OpenAI-wire and Anthropic-wire endpoints are supported. Heavy Claude users get the 1-hour cross-session cache discount when using the native Anthropic path.

​
Feature Cheat Sheet

Terminal UI

Full TUI: multiline editing, slash-command completion, conversation history, streaming tool output

Messaging Gateway

Run hermes gateway setup to bind bot tokens and chat from any IM platform

Skills System

Procedural memory + Skills Hub (agentskills.io) — gets smarter as you use it

MCP Integration

Plug in any MCP server, including community Linux desktop-control MCP

Scheduled Tasks

Built-in cron — natural language like “send me a daily report at 9am”

Subagents

Spawn isolated subagents for parallel work; call tools via RPC from Python scripts

​
Migrating from OpenClaw

If you’re coming from OpenClaw, Hermes includes a built-in migration tool: 
hermes claw migrate              # Interactive full migration
hermes claw migrate --dry-run    # Preview what would be migrated
hermes claw migrate --preset user-data   # User data only, no secrets
hermes claw migrate --overwrite  # Overwrite conflicts
It imports SOUL.md, memories (MEMORY.md / USER.md), user-created skills, command allowlists, messaging platform configs, API keys (Telegram / OpenRouter / OpenAI / Anthropic / ElevenLabs), TTS assets, and workspace instructions.

​
FAQ

  • Hermes Agent: Python implementation by Nous Research — focused on the self-improving loop with skill evolution and cross-session memory, research-friendly (supports trajectory generation)
  • OpenClaw: Node.js implementation — focused on local privacy + multi-IM-platform interplay
  • FastClaw: Go single binary — focused on multi-agent management with a Dashboard form factor
All three can reach APIYI’s full model matrix. Pick whichever fits your use case.
Yes. Hermes supports both the OpenAI-compatible and Anthropic-native protocols:
  • OpenAI endpoint: https://api.apiyi.com/v1 — full model matrix
  • Anthropic endpoint: https://api.apiyi.com (no /v1) — Claude family
Use the model IDs from APIYI’s docs directly (e.g. gpt-5.4, claude-sonnet-4-6, deepseek-v3.2, gemini-3.1-pro-preview). Claude-heavy users should go through the Anthropic-native path to unlock Hermes’s 1-hour cross-session prompt cache.
Hermes runs a single Gateway process that manages bot connections across Telegram / Discord / Slack / WhatsApp / Signal. hermes gateway setup walks you through pasting tokens, and hermes gateway start routes messages from every platform into the same agent instance — conversations stay continuous across platforms, so you can pick up a Telegram thread on Discord.It includes voice-memo transcription, and cron deliveries flow through the same gateway.
Yes — and Hermes is built for that. It offers seven terminal backends:
  • Local / Docker / SSH / Singularity: traditional deployments
  • Modal / Daytona: serverless persistence — hibernates when idle, wakes on demand, nearly zero cost between sessions
  • Vercel Sandbox: edge runtime
A $5 VPS is enough to stand it up 24/7. Sending tasks from your phone via Telegram to a cloud VM works out of the box.

Project Repository

github.com/NousResearch/hermes-agent

Official Docs

hermes-agent.nousresearch.com/docs/

OpenClaw Alternative

For the local-privacy + IM interplay use case

FastClaw Alternative

For the multi-agent factory + Dashboard use case