AI Inference Benchmark: Keln vs Fireworks, Baseten, Modal, DeepInfra & Z.ai
GuideHermes Agent4 min read

Run Hermes Agent on Keln

Keln is now in the Hermes Agent plugin catalog. Install it with one command and run your agent on fast, verified open models.

What is Hermes Agent?

Hermes Agent logo

Hermes Agent is an open-source AI agent built by Nous Research. It runs on your computer or server, remembers previous sessions and creates reusable skills as it works. It's MIT-licensed and supports multiple model providers.

Unlike a regular chatbot, Hermes can keep working after you close the chat. Here's what you can do with it:

  • Run it on your own machine. Hermes works on a laptop, VPS or cloud VM. You can message it from the terminal, Telegram, Discord, Slack, WhatsApp or Signal.
  • Schedule tasks. Ask Hermes to prepare a daily report or run a weekly audit. It will run the job on schedule.
  • Run tasks in parallel. Hermes can start subagents to work on different parts of a task at once.
  • Use built-in tools. It has more than 40 tools, including web search, browser automation and vision.

How popular is Hermes Agent?

Hermes Agent is one of the most popular open-source AI agents, with more than 262,000 GitHub stars and 58,000 forks.

How to run Hermes Agent on Keln

You'll need Hermes Agent 0.22.6 or newer and a Keln API key.

1. Install the Keln provider.

Run:

hermes plugins install keln-provider

Hermes will ask for your Keln API key and whether to enable the plugin. Enter y to enable it.

2. Choose Keln and a model.

hermes model

Pick Keln from the list. Hermes then asks about your key and the URL. Press Enter on both. Last, pick a model. Don't skip this part, or Hermes keeps using its old model.

3. Start chatting.

hermes chat

Got stuck? The Hermes Agent setup guide shows every step.

Why run Hermes Agent on Keln?

A single agent task can make hundreds of model calls. If the model is slow, requests fail or there is no capacity, the task takes longer or stops.

Keln runs open models and manages all the infrastructure for them. It does more than a single provider does. Here are the things that matter most when you run agents, and how Keln helps with each.

1. Consistent speed

A few seconds of delay might not matter in a chat. An agent can take hundreds of steps to complete a task, so the delay grows to minutes.

Keln targets under 2 seconds to the first token and at least 50 tokens per second for every model. Keln is open about its numbers: you can check each model's speed over the last 24 hours on its model page.

2. Capacity when you need it

Hermes can run scheduled jobs overnight or launch several subagents at once. You shouldn't have to plan capacity for that.

Keln scales with your workload, without capacity reservations or manual planning. Your agents won't run into the rate limits you may hit with a single provider.

3. Full model quality and reliable tool calling

Hermes does its work through tool calls, such as a web search or a browser action. When a model returns a wrong tool call, that step fails.

Keln continuously benchmarks the models it serves, including their tool calling, and removes degraded instances. Keln makes sure you get the full model on every request.

4. Long context and prompt caching

Hermes sends the conversation history to the model with each request. As the task gets longer, so does the context.

GLM 5.3, Kimi K3 and DeepSeek V4.1 Flash support up to 1 million tokens of context on Keln.

Prompt caching avoids processing the same context repeatedly, reducing latency and cost. In our agent benchmark, Keln achieved a 94% cache hit rate on DeepSeek V4.1 Flash.

5. Zero data retention

Hermes may work with your files, messages and other personal data. Keln does not store the prompts and outputs your agent sends, and does not train on them. It keeps only request metadata, which you can check on your Usage page.

Get a Keln API key

Which Keln models work best in Hermes Agent?

Run hermes model to see all the Keln chat models you can use. Here are a few we recommend to start with:

UseModels
Complex, multi-step tasksGLM 5.3, Kimi K3
Fast, high-volume tasksGLM 5.3 Flash, DeepSeek V4.1 Flash

You can switch models anytime with hermes model. See all Keln models in the model catalog.

Get started

Get your Keln API key and follow the Hermes Agent setup guide to connect your agent.

Where to go next

FAQ

Hermes Agent is an open-source AI agent developed by Nous Research. It runs on your computer or server, remembers past sessions and creates reusable skills from past tasks.

Yes. Hermes Agent is free and open source under the MIT license. You still pay for the tokens you use through your model provider.

Run hermes plugins install keln-provider and enter your Keln API key. Then run hermes model and choose Keln.

Hermes works with every chat model in the Keln catalog. GLM 5.3, Kimi K3, Qwen3.8 Max and DeepSeek V4.1 Flash are a few examples.

No. Keln doesn't store your prompts or outputs and doesn't train on your data. It keeps only request metadata, which you can also check on your Usage page.

Hermes AgentGuideGLM 5.3Kimi K3DeepSeek V4.1 Flash

Run GLM, Kimi, and DeepSeek in Hermes Agent

Connect Hermes Agent to Keln to run open models with fast inference, reliable performance, and simple per-token pricing.