Does Hermes Agent support Claude?

Yes. Self-hosted Hermes Agent has a built-in Anthropic provider that accepts ANTHROPIC_API_KEY, ANTHROPIC_TOKEN, or CLAUDE_CODE_OAUTH_TOKEN, and works directly or via OpenRouter. On a paid OpenClaw Launch plan, pick Claude Sonnet 5, Opus 5, or Haiku 4.5 from the model dropdown, using the included credits or your own Anthropic key; Claude subscription sign-in is not offered for hosted bots.

Which Claude model should I use with Hermes?

Claude Sonnet 5 is the recommended default — best balance of speed, quality, and cost. Use Opus 5 for deep research and long-horizon reasoning. Use Haiku 4.5 for fast, high-volume chat.

Can I switch between Claude models without restarting Hermes?

Yes. Use the /model command in any connected channel — for example /model anthropic/claude-opus-4.6 — to switch the active model instantly without restarting the gateway.

Do I need a separate Anthropic API key for Hermes on OpenClaw Launch?

No. OpenClaw Launch includes AI credits that cover Claude requests via OpenRouter. Bring your own Anthropic key only if you want billing under your own Anthropic account.

Which environment variables does Hermes read for Anthropic?

The built-in anthropic provider accepts ANTHROPIC_API_KEY, ANTHROPIC_TOKEN, or CLAUDE_CODE_OAUTH_TOKEN. Set one, then run hermes model or set provider: anthropic and a default model in /opt/data/config.yaml.

Should I use the Anthropic API or OpenRouter with Hermes?

Use the Anthropic API direct for lowest latency if you only run Claude; rate limits are Anthropic tier-based. Use OpenRouter for one key across 200+ models, useful when you switch between Claude, GPT, Gemini, DeepSeek, and others through Hermes /model.

Can I bring my own Anthropic key to OpenClaw Launch Hermes?

Yes. Managed Hermes deploys support BYOK: paste your Anthropic or OpenRouter key on the API Keys page, pick a Claude model, and its requests run through your account.

← Home

Guide

Hermes Agent + Claude: Use Anthropic's Claude Models with Hermes

Claude — Anthropic's model family — is one of the strongest choices for Hermes Agent. Opus 5 handles deep research and long-running tasks, Sonnet 5 is the everyday workhorse, and Haiku 4.5 delivers fast, low-cost responses.

What Is Claude?

Claude is Anthropic's family of large language models, known for nuanced reasoning, precise instruction-following, strong writing quality, and honest, helpful behavior. Unlike some competitors that optimize primarily for benchmark scores, Claude is designed with safety and usefulness together — qualities that matter a lot for agent workloads where the model makes decisions autonomously.

Hermes Agent supports Claude through two paths: the Anthropic API directly (via ANTHROPIC_API_KEY) or via OpenRouter (which routes to Claude with a single key that covers 200+ other models too).

Claude Model Lineup for Hermes

Model	Best For	Context	Cost (Input)
Claude Opus 5	Complex reasoning, research, long-running agent tasks	1M tokens	~$5/M tokens
Claude Sonnet 5	Best balance of speed and quality — recommended default	1M tokens	~$2/M tokens
Claude Haiku 4.5	Fast responses, daily chat, high-volume messaging	200K tokens	~$1/M tokens
Claude Opus 4.6 / Sonnet 4.6	Previous generation — still available if you want to stay on them	1M tokens	~$5 / ~$3/M tokens

For most Hermes users, Claude Sonnet 5 is the right starting point. It handles tool calls reliably, produces high-quality responses at conversational speed, and costs roughly $2 per million input tokens. Upgrade to Opus when you need its reasoning depth; drop to Haiku when volume and speed matter more than depth.

Option 1: Hermes Agent on OpenClaw Launch (Easiest)

The fastest path to a Claude-powered Hermes Agent. No API key required, no Docker setup, no config file editing.

Go to openclawlaunch.com/hermes-hosting and start a Hermes deploy.
Select Claude Sonnet 5 from the model dropdown — or Opus 5 for the hardest tasks and Haiku 4.5 for speed. Sonnet 4.6 and Opus 4.6 are still listed if you want to stay on them.
Connect your messaging channel — Telegram, Discord, WhatsApp, or others.
Click Deploy. Your Claude-powered Hermes Agent is live in roughly 30 seconds.

Tip: OpenClaw Launch routes Claude requests through OpenRouter automatically. AI credits are included in every paid Hermes plan — no separate Anthropic billing required unless you bring your own key. The 30-minute free trial runs on the free trial models; Claude needs a paid plan.

Option 2: Anthropic API Direct (Self-Hosted)

If you're running Hermes on your own server with a direct Anthropic API key, set the environment variable and tell Hermes to use the anthropic provider:

# Hermes reads ANTHROPIC_API_KEY, ANTHROPIC_TOKEN,
# or CLAUDE_CODE_OAUTH_TOKEN — all three are accepted
export ANTHROPIC_API_KEY=sk-ant-...

# Set provider and default model via CLI
# Pick the provider and default model (interactive wizard)
hermes model

# Or configure /opt/data/config.yaml directly:
# model:
#   provider: anthropic
#   default: claude-sonnet-4.6

Get your API key from the Anthropic Console. Usage-based billing, no monthly minimum.

Option 3: Claude via OpenRouter (Self-Hosted)

OpenRouter lets you reach every Claude model with a single key, alongside GPT, Gemini, DeepSeek, Grok, and 200+ others. Useful when you want the flexibility to switch models without managing multiple API keys.

export OPENROUTER_API_KEY=sk-or-...

# Pick the provider and default model (interactive wizard)
hermes model

# config.yaml equivalent:
# model:
#   provider: openrouter
#   default: anthropic/claude-sonnet-4.6

When to Choose Each Claude Model

Choose Claude Sonnet 5 for everyday agent use: answering questions, writing, research, coding, and multi-step tool calls. It handles complex requests well without the cost of Opus.

Choose Claude Opus 5 for demanding tasks that require sustained deep reasoning: long-horizon planning, multi-document analysis, complex code architecture decisions. Its 1M context window makes it especially powerful for tasks that involve large bodies of text.

Choose Claude Haiku 4.5 when speed and cost per message matter most — support bots with high message volume, quick Q&A, or fast triage workflows where responses don't need maximum depth.

Switching Models at Runtime

Hermes supports a built-in /model command in chat that switches the active model without restarting the gateway. From any connected channel:

/model anthropic/claude-haiku-4.5
/model anthropic/claude-sonnet-4.6
/model anthropic/claude-opus-4.6

On OpenClaw Launch, the picker lists all curated Claude options alongside GPT, Gemini, Grok, DeepSeek, GLM, Kimi, MiniMax, and others.

BYOK on OpenClaw Launch

On managed OpenClaw Launch deploys, you can use your own Anthropic or OpenRouter key instead of bundled AI credits. Paste it on the API Keys page, then pick a Claude model for your bot — its requests then run through your account, with usage and billing under your direct control.

Self-hosted Hermes also reads a CLAUDE_CODE_OAUTH_TOKEN env var. OpenClaw Launch does not offer Claude Pro/Max subscription sign-in for hosted bots — see Hermes with a Claude subscription for why, and for the ChatGPT-subscription and Anthropic-key routes that do work.

Claude Direct vs OpenRouter for Hermes

Feature	Anthropic Direct	OpenRouter
Latency	Lowest (one hop)	Slightly higher (proxy hop)
Models available	Anthropic Claude family only	200+ models across all major providers
Billing	Anthropic Console	Unified OpenRouter billing
Switching via `/model`	Need separate keys for other providers	One key covers everything

Use the Anthropic API direct if you stay on Claude long-term and want the lowest latency. Use OpenRouter if you want to switch freely between Claude, GPT, Gemini, DeepSeek, and others through Hermes's /model command.

Hermes Anthropic Setup: Provider, API Key, and Auth Paths

Searching for "Hermes Anthropic"? It is the same thing as Hermes with Claude: Anthropic builds the Claude models, and Hermes Agent (built by Nous Research) has a built-in anthropic inference provider for them. Hermes reaches Anthropic through three paths: the Anthropic API directly, OpenRouter, or zero configuration on OpenClaw Launch.

Create a key. Generate an API key in the Anthropic Console. Billing is usage-based with no monthly minimum.
Export it. The anthropic provider accepts any of ANTHROPIC_API_KEY, ANTHROPIC_TOKEN, or CLAUDE_CODE_OAUTH_TOKEN.
Pick the provider and model. Run hermes model (interactive wizard) or set provider: anthropic and a default model in /opt/data/config.yaml, as shown in Option 2 above.
Or route through OpenRouter. Set OPENROUTER_API_KEY with provider: openrouter and a model id like anthropic/claude-sonnet-4.6. One key then covers Claude, GPT, Gemini, DeepSeek, Grok, and 200+ others.

One extra row for the direct-vs-OpenRouter comparison above: with Anthropic direct, rate limits are Anthropic tier-based; with OpenRouter they are OpenRouter managed. On managed OpenClaw Launch deploys you can also bring your own Anthropic or OpenRouter key (BYOK) instead of the bundled AI credits — see the BYOK section above.

What's Next?

Anthropic API key setup — Provider, env vars, and direct-vs-OpenRouter trade-offs on this page
Hermes Agent + OpenRouter — One key for 200+ models, including Claude, GPT, and Gemini
Hermes Agent + Telegram — Connect your Claude-powered Hermes bot to Telegram
Hermes Agent + MCP — Extend your Claude agent with Model Context Protocol servers