> ## Documentation Index > Fetch the complete documentation index at: https://concentrate.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # Hermes Agent (Beta) > Use models on Concentrate with Nous Research's Hermes Agent > [!NOTE] > > Hermes integration is currently in **beta**. Some features may not work, some may only work with certain models. ## Why Use Concentrate with Hermes? [Hermes Agent](https://github.com/NousResearch/hermes-agent) is an open-source agent from Nous Research that runs in your terminal, with tools for the shell, files, the web, MCP servers, subagents and computer use. It accepts any OpenAI- or Anthropic-compatible API as a custom provider. By routing Hermes through Concentrate, you use models from every provider in our [catalog](https://concentrate.ai/models) with one API key, and get centralized visibility and control over your team's usage: spend tracking, budget limits, and everything visible in your Concentrate dashboard in real time. ## What Works | Feature | Status | Notes | | - | - | - | | **Chat and coding** | Supported | All models on Concentrate | | **Tool calls** | Supported | Including several tool calls at once. Needs a model that supports tool use | | **MCP servers** | Supported | Tools from MCP servers in your Hermes config | | **Subagents** | Supported | `delegate_task` runs child agents in parallel | | **Images** | Supported | Pasted images and images read by tools. Needs a model that accepts images | | **Computer use** | Supported | Hermes' `computer_use` tool, with screenshots sent back to the model. Needs a model that accepts images | | **Sessions** | Supported | Resume with `hermes chat --continue` or `--resume ` | | **Switching models** | Supported | Switch with `/model` mid-session, including across providers | ## Prerequisites * A Concentrate AI account with an active API key (starts with `sk-cn`) * [Hermes Agent](https://github.com/NousResearch/hermes-agent#quick-install) installed: ```bash Bash theme={null} curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash ``` ## Quick Setup ```bash Bash theme={null} curl -fsSL https://concentrate.ai/scripts/setup-hermes.sh | bash ``` The script saves your API key to Hermes' own `~/.hermes/.env` as `CONCENTRATE_HERMES_API_KEY`, so it doesn't change your shell profile, and Hermes can use a different key from your other tools. If you already have `CONCENTRATE_API_KEY` set, it asks whether to reuse it. It adds a `concentrate` provider to `~/.hermes/config.yaml` that uses our Responses API (for the other API formats, see [Other API formats](#other-api-formats)), and makes Concentrate your default provider, asking first if you already have one. Your other settings are kept, and `config.yaml` is backed up first. Without a terminal (for example in CI), the script can't ask, so it keeps any default provider you already have, and uses a saved key or `CONCENTRATE_API_KEY`. If neither is set, pass the key with `--key `. Then start Hermes in a project directory: ```bash theme={null} hermes ``` ## Manual Setup If you prefer to configure Hermes manually or the script doesn't work for your environment: **1. Save your API key** Add this line to Hermes' `.env` file, `~/.hermes/.env` (create it if it doesn't exist): ```ini theme={null} CONCENTRATE_HERMES_API_KEY=sk-cn-v1-... ``` Editing the file keeps the key out of your shell history. **2. Add Concentrate to \~/.hermes/config.yaml** Add a `concentrate` provider and make it the default. Keep any other settings already in the file. `key_env` is the name of the variable, not the key itself: ```yaml theme={null} model: provider: concentrate default: anthropic/claude-sonnet-5-5 providers: concentrate: name: Concentrate base_url: https://api.concentrate.ai/v1 key_env: CONCENTRATE_HERMES_API_KEY transport: codex_responses ``` `codex_responses` makes Hermes use our [Responses API](https://concentrate.ai/docs/api-reference/endpoint/create-response.md). Use any model from the [supported models](https://concentrate.ai/docs/api-reference/endpoint/supported-models.md) page with its `provider/model` ID. **3. Verify the connection** Send a one-off prompt, then start Hermes in a project directory: ```bash theme={null} hermes chat --oneshot -q "Reply with exactly: hermes ok" cd /path/to/project hermes ``` ## Switch models Inside a session, name the model and the provider: ```text theme={null} /model openai/gpt-6-luna --provider concentrate ``` To start a session on a different model, pass it on the command line: ```bash theme={null} hermes chat --provider concentrate -m ai-studio/gemini-3.8-flash ``` > [!WARNING] > > Don't use the `provider:model` form (for example `/model concentrate:openai/gpt-6-luna`) with a custom provider. Hermes sends the whole string as the model ID, which Concentrate rejects. ## Other API formats We recommend the Responses API (`codex_responses`) shown above. It gets the best prompt-cache hit rate, which keeps long agent sessions cheaper. Hermes can also use our Chat Completions and Messages APIs, which are in beta. To use them, add them under `providers:` in the same file: ```yaml theme={null} concentrate-chat: name: Concentrate (Chat Completions) base_url: https://api.concentrate.ai/v1 key_env: CONCENTRATE_HERMES_API_KEY transport: chat_completions concentrate-messages: name: Concentrate (Messages) base_url: https://api.concentrate.ai key_env: CONCENTRATE_HERMES_API_KEY transport: anthropic_messages ``` Then select one with `--provider`, for example `hermes chat --provider concentrate-messages -m anthropic/claude-haiku-4-5`. > [!WARNING] > > Claude models on Chat Completions don't use prompt caching yet, so every turn pays the full input price. In a long agent session that can cost several times more than Responses or Messages. Use Responses or Messages for Claude. ## Tips * **Context window.** Hermes needs a model with a context window of at least 64K tokens, because its tools and system prompt take up a large part of every request. Check a model's context window on the [supported models](https://concentrate.ai/docs/api-reference/endpoint/supported-models.md) page. * **Images on text-only models.** If you send an image to a model that can't read images, Hermes first describes it with a vision model and passes the description on. That extra call is billed like any other request. * **Side requests.** Besides your conversation, Hermes makes short side requests, for example to title sessions or describe images. They appear in your Concentrate dashboard alongside the main requests. * **Costs.** The Concentrate dashboard is the source of truth for spend. ## Remove Concentrate AI ```bash Bash theme={null} curl -fsSL https://concentrate.ai/scripts/uninstall-hermes.sh | bash ``` This removes the `concentrate` provider and `CONCENTRATE_HERMES_API_KEY`. Remove any providers you added by hand yourself. Other settings are kept. If Concentrate was your default provider, run `hermes model` afterwards to choose a new one. ## Troubleshooting > **401 Invalid API key** > > 1. Check the key is saved: `hermes config get CONCENTRATE_HERMES_API_KEY` shows its first and last characters. To replace it, re-run the setup script or edit the line in `~/.hermes/.env`. > 2. Check `key_env` in `config.yaml` is the variable's name (`CONCENTRATE_HERMES_API_KEY`), not the key itself. > **400 Invalid request: model** > > The model ID isn't one Concentrate knows. Use the exact `provider/model` ID from the [supported models](https://concentrate.ai/docs/api-reference/endpoint/supported-models.md) page, and switch with `/model --provider concentrate`, not `concentrate:`. > **402 Insufficient funds** > > Your balance is empty. Add funds on the [billing page](https://concentrate.ai/personal/billing).