The endpoint thatnever logs your agent.

Private, tool-calling models forHermes Agent.

Setup walkthrough

Video coming soon

Built for agents,
private by default

Get an API key Jump to setup

Hermes Agent

One endpoint, the whole catalog

Hermes Agent already speaks the OpenAI API, so one custom endpoint and one key reach every tool-calling model here, from private open-weight agents to the frontier names you know.

Do you keep what my agent sends?

Kimi K2.6

Not on the router. Prompts, tool results, and files pass through AnonRouter in memory and are never written to disk, and on a Private route like this one the provider keeps nothing once the reply is sent.

Private

Give Hermes a task...

No agent logs on our side

Every request is forwarded in memory and billing is settled from token counts, so nothing your agent reads or writes sits in a log here to leak, sell, or train on.

STEP 01

Create an API key

Sign up or sign in, open API keys, and create a key with OpenAI-compatible access ticked. Hermes sends the same key with every request, and that box is what lets it through. Give it a credit limit: an agent that runs on its own stops at the cap instead of draining your balance. The key starts with ar_ and is shown once.

Create a key

API keysNew key

Search by name or prefix...

11 active

NameKeyUsageLimit

hermes-agentar_vhwHUTw...$6.18$20.00

hermes-gatewayar_I5RwgEY...$2.73$10.00

research-botar_Xr1ICrc...$1.06$3.00

inbox-triagear_wGpcFAE...$0.42$2.00

local-devar_sxENWfw...$0.087$1.00

side-projectar_gv4l0DE...$0$2.00

Create API key

STEP 02

Choose your model

Every model here can call tools and has at least the 64K context Hermes needs for its system prompt, tools, and working state. Start from a recommended one or search the rest.

149 models

Recommended for agents

Claude Sonnet 5anthropic/claude-sonnet-5AnonymousStrong at long, multi-step tool use, with a 1M token context.Anonymousby anthropic|1M context|$3.00/M input|$15.00/M outputGLM 5.2z-ai/glm-5.2PrivatePrivate routes and a 1M token context for long agent runs.Privateby z-ai|524K context|$1.75/M input|$5.75/M outputKimi K2.6moonshotai/kimi-k2.6PrivateOpen weights built for agent work, on Private routes.Privateby moonshotai|256K context|$0.75/M input|$3.50/M outputDeepSeek V4 Flashdeepseek/deepseek-v4-flashPrivateCheap enough to leave running, with a 1M token context.Privateby deepseek|via DeepInfra|1M context|$0.09/M input|$0.18/M outputVenice Uncensored 1.2Uncensoredvenice/venice-uncensored-1.2PrivateUnfiltered, on Private routes, at a very low price.Privateby venice|128K context|$0.20/M input|$0.90/M output

Explore hundreds more models

Where a card names a provider, its label and price are that default route; otherwise the label is the least private route a message could take. The router tries the most private route first, and a minimum privacy level in your privacy settings stops it falling back below that. Private routes keep nothing after the request; on Anonymous routes the provider sees the prompt but not who sent it.

STEP 03

Copy your connection details

Paste your key into the box and it drops into the details below. Nothing is sent anywhere. These are Hermes' setup prompts, word for word and in order, with the answer to each.

Your API key (optional)

Never leaves this tab. It is masked below and only goes to your clipboard.

hermes model

Select provider:
Custom endpoint (enter URL manually)
Select
API base URL:
https://api.anonrouter.ai/v1
API key [optional]:
ar_REPLACE_WITH_YOUR_API_KEY
Select API compatibility mode:
Auto-detect (press Enter)
Select
Select model [1-N] or type name:
anthropic/claude-sonnet-5
Context length in tokens:
1000000
Display name:
AnonRouter
Select reasoning effort:
Disable reasoning
Select

STEP 04

Connect in Hermes Agent

The rest happens in your terminal. The wizard checks the endpoint as you go and saves everything to Hermes' own config, key included, so there is nothing to edit by hand.

Install Hermes Agent: curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bashOn Windows, in PowerShell: iex (irm https://hermes-agent.nousresearch.com/install.ps1). Skip this if you have it.
Run hermes model
At Select provider: choose Custom endpoint (enter URL manually)
Answer each prompt with the values abovePress Enter at the compatibility mode, and type the model name when it lists models.
At Select reasoning effort: choose Disable reasoningHermes asks for medium by default, which models without adjustable reasoning turn down.
Run hermes and give it a taskRun hermes model again whenever you want to switch models.

Or hand the whole thing to your agent

One prompt carries the entire setup. Give it to your agent, review what it configures, and you are done.

OR IN THE CONFIG FILE

Rather edit the files yourself?

Replace the model section of ~/.hermes/config.yaml with this one and set agent.reasoning_effort to none, then put the key in ~/.hermes/.env. Switch models mid-session with /model custom:anthropic/claude-sonnet-5.

~/.hermes/config.yaml

model:
  provider: custom
  base_url: https://api.anonrouter.ai/v1
  default: "anthropic/claude-sonnet-5"
  key_env: ANONROUTER_API_KEY
  context_length: 1000000
 
agent:
  reasoning_effort: none

~/.hermes/.env

ANONROUTER_API_KEY=ar_REPLACE_WITH_YOUR_API_KEY

Compatibility mode

Hermes Agent sends the same key with every message, so these requests use the compatibility broker: the key is checked, a single-use ticket is minted in memory, and the request is forwarded with metadata-only logging. AnonRouter still logs none of your messages, but this flow is not unlinkable the way the ticket flow is. Read the docs

Why AnonRouter

Hermes sends everything it is working on. We keep none of it.

An agent ships its prompt, tool schemas, tool results, and the files it reads with every request. AnonRouter forwards it all in memory and settles billing from usage metadata alone.

No agent logs

Prompts, tool results, files, and responses pass through in memory and are never written to disk. There is no request history on our side to leak, sell, or train on.

Labeled privacy

Every route is labeled with the guarantee it actually provides, so an agent reading your files or inbox can stay on Private routes.

A cap per key

Prepaid credits and a total credit limit on each key, so an agent running unattended stops at the cap instead of draining your balance.

Troubleshooting

If a message fails

Most endpoint errors come with a status code. Match it below, or read the full Hermes Agent guide in the docs.

401

Key not accepted

The key is missing, mistyped, or revoked. Paste the whole key, starting with ar_. With the config-file route, check that ANONROUTER_API_KEY in ~/.hermes/.env matches key_env exactly.

403

Compatibility is off

This key does not have OpenAI-compatible access. Turn it on for the key in API keys, or create a new key with it ticked.

404

Wrong URL or model

The base URL is https://api.anonrouter.ai/v1; Hermes adds /chat/completions itself. The model has to be an exact id from this page, not a display name.

Refused

Hermes rejects the model

Hermes refuses models under 64,000 tokens of context, and guesses when it cannot read the size. Pick a model from this page and enter its context length when the wizard asks.

400

Reasoning turned down

Hermes asks for medium reasoning by default, and a model without adjustable reasoning turns that down. Run hermes model and choose Disable reasoning, or set agent.reasoning_effort to none.

402

Out of credits

Your balance is empty, or the key hit its credit limit. Check what the agent was doing before raising the limit, then top up or switch to a cheaper model.

Still stuck? Ask us on DiscordTell us the status code and the model you picked. Never post your API key there.Join our Discord

Hermes Agent questions

There is no setup fee and no subscription. Requests are paid from prepaid AnonRouter credits at the rate of the model you pick, and a total credit limit on the key caps what one agent can spend.

Not yet. A key has a total credit limit, which works as a circuit breaker; it does not reset each day or month. For unattended runs, pair it with Hermes' own tool-loop guardrails.

Usage metadata only: model, token counts, status, latency, and cost. Prompts, tool results, files, and responses pass through in memory and are never written to disk. The model's provider handles each request according to the privacy label on the model you picked.

In ~/.hermes/.env. The wizard saves it there under a name it builds from the endpoint; the config-file route uses ANONROUTER_API_KEY. Keep ~/.hermes private, and never commit it.

Private: Prompt and response content is not persistently retained after the request. Anonymous: Your identity is hidden, but the inference provider can see the prompt.

Hermes asks every custom endpoint for medium reasoning unless told otherwise, and AnonRouter turns that down for models without adjustable reasoning. Off works with every model. If yours supports reasoning effort, run hermes model again and pick a level.

You can, but a pinned model is more predictable for an agent: Hermes needs tool calling and a large context on every request, and a model from this page is known to have both.

No. AnonRouter is an independent service and is not affiliated with Nous Research or Hermes Agent. Hermes works with any OpenAI-compatible endpoint, and AnonRouter is one.

What your agent does is not our business.

Get an API key

AnonRouter is an independent service and is not affiliated with Hermes Agent.