> ## Documentation Index
> Fetch the complete documentation index at: https://docs.powertokens.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Hermes Agent setup guide

> Connect Hermes Agent to the powertokens platform through an OpenAI-compatible endpoint.

Hermes Agent can use custom OpenAI-compatible endpoints. For the `powertokens platform`, the stable path is the platform `/v1/chat/completions` compatible endpoint.

The platform `/v1/messages` Anthropic Messages endpoint is available, but Hermes Agent provider routing differs by version. In Hermes Agent v0.10.0, the main chat path may still request `/chat/completions` even when `api_mode: anthropic_messages` is configured, and may normalize `glm-5.2` to `glm-5-2`. This page treats `anthropic_messages` as a version-gated setup direction, not as a stable Hermes Agent promise.

If you need to choose the endpoint first, see [Text model protocols and endpoints](/en/ecosystem-tools/text-model-protocols).

## Choose a path

| Goal                               | Hermes setup                                                                                             | Base URL                        | Model scope                                                                                             |
| ---------------------------------- | -------------------------------------------------------------------------------------------------------- | ------------------------------- | ------------------------------------------------------------------------------------------------------- |
| Quick text-model setup             | Custom endpoint                                                                                          | `https://api.powertokens.ai/v1` | OpenAI-compatible text models                                                                           |
| Anthropic Messages verification    | Use `api_mode: anthropic_messages` only after your Hermes version is confirmed to request `/v1/messages` | `https://api.powertokens.ai`    | Recommended Zhipu, MiniMax, Qwen, BytePlus DeepSeek, and re-validated Seed models that support Messages |
| OpenAI-compatible preferred models | Custom endpoint                                                                                          | `https://api.powertokens.ai/v1` | Text models whose upstream provider does not support `/v1/messages`, such as `seed-1-6-flash-250715`    |

## OpenAI-compatible endpoint

Start Hermes model setup:

```bash theme={null}
hermes model
```

Use these values:

| Field         | Value                                                         |
| ------------- | ------------------------------------------------------------- |
| Endpoint type | Custom endpoint or OpenAI-compatible endpoint                 |
| URL           | `https://api.powertokens.ai/v1`                               |
| API key       | Your `powertokens platform` API key                           |
| Model         | For example `glm-5.2`, `MiniMax-M3`, or `seed-2-0-pro-260328` |

If your Hermes version asks for the full chat completions endpoint, use:

```text theme={null}
https://api.powertokens.ai/v1/chat/completions
```

## anthropic\_messages version check

Only use `anthropic_messages` when your Hermes version is confirmed to send the request to `POST /v1/messages`. Adapt this field shape to the config format used by your Hermes version:

```yaml theme={null}
provider: powertokens-anthropic
api_mode: anthropic_messages
base_url: https://api.powertokens.ai
api_key: ${POWERTOKENS_API_KEY}
model: glm-5.2
```

After configuring it, inspect the Hermes request dump or debug log. The final request URL should be `https://api.powertokens.ai/v1/messages`. If the request still lands on `/chat/completions`, do not treat this path as a stable Hermes Agent setup.

## Recommended models

| Path                               | Suggested model        | Notes                                                  |
| ---------------------------------- | ---------------------- | ------------------------------------------------------ |
| OpenAI-compatible                  | `glm-5.2`              | Default coding model                                   |
| OpenAI-compatible                  | `seed-2-0-pro-260328`  | BytePlus text path                                     |
| `anthropic_messages` version check | `glm-5.2`              | Use only after Hermes actually requests `/v1/messages` |
| `anthropic_messages` version check | `MiniMax-M3`           | Use only after Hermes actually requests `/v1/messages` |
| `anthropic_messages` version check | `deepseek-v3-2-251201` | Use only after Hermes actually requests `/v1/messages` |

Recommended Anthropic tool-workflow model IDs:

```text theme={null}
glm-4.5-air
glm-4.7
glm-4.7-flash
glm-5
glm-5-turbo
glm-5.2
MiniMax-M2.5
MiniMax-M2.5-highspeed
MiniMax-M2.7
MiniMax-M2.7-highspeed
MiniMax-M3
qwen3-coder-plus
qwen3-max
qwen3.5-flash
qwen3.6-plus
deepseek-v3-2-251201
seed-1-6-250915
seed-1-8-251228
seed-2-0-lite-260228
seed-2-0-mini-260215
seed-2-0-pro-260328
```

## Verify the OpenAI-compatible path

```bash theme={null}
curl -X POST "https://api.powertokens.ai/v1/chat/completions" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_POWERTOKENS_API_KEY" \
  --data '{
    "model": "glm-5.2",
    "messages": [
      {"role": "user", "content": "Reply with ok."}
    ],
    "stream": false
  }'
```

## Verify the anthropic\_messages path

```bash theme={null}
curl -X POST "https://api.powertokens.ai/v1/messages" \
  -H "Content-Type: application/json" \
  -H "x-api-key: YOUR_POWERTOKENS_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  --data '{
    "model": "MiniMax-M3",
    "max_tokens": 64,
    "messages": [
      {"role": "user", "content": "Reply with ok."}
    ],
    "stream": false
  }'
```

## Boundaries

* The Hermes custom endpoint path reliably covers `/v1/chat/completions`; it does not prove the main chat path is using Anthropic Messages.
* `api_mode: anthropic_messages` should only be used when the current Hermes version is verified to request `/v1/messages`.
* Tool execution, terminal control, long context, caching, and streaming should be validated with real Hermes tasks.
* If the `anthropic_messages` path is unstable in real tasks, switch to the OpenAI-compatible endpoint and keep the failed request id.
* Do not copy MiniMax upstream-only fields into platform examples; this page describes the public unified API contract.