---
title: How Hermes Agent Uses Grok Through xAI
description: "Trace the path from an xAI API key to a Hermes Agent run and learn which layer owns authentication, execution, and Telegram delivery."
canonical: "https://deploy-hermes.com/technical/how-hermes-agent-uses-grok"
last-updated: "2026-08-23"
---

# How Hermes Agent Uses Grok Through xAI

> Trace the path from an xAI API key to a Hermes Agent run and learn which layer owns authentication, execution, and Telegram delivery.

Canonical: https://deploy-hermes.com/technical/how-hermes-agent-uses-grok
Updated: 2026-08-23 · Search intent: informational
Category: [Technical](/technical)

Hermes Agent uses Grok through xAI as a provider-backed runtime path. DeployHermes validates the xAI key, seals it, maps it to the runtime's xAI credential name, and supplies it when the selected agent claims a run.

## Credential connection

A workspace operator pastes an xAI API key into provider settings. The backend validates it against xAI's models endpoint before sealing it. The browser receives an account label rather than the secret, and later status queries expose connection state without returning plaintext.

This is deliberately different from a model benchmark. It proves that the credential can authenticate at connection time; a run still has to pass policy and runtime checks.

## Run dispatch

When an agent run is claimed, the runtime receives the connected provider credential for the permitted provider policy. The xAI value is mapped to `XAI_API_KEY`, which is the environment name the runtime uses for xAI authentication. The agent then makes the provider request through its normal execution path.

The runtime owns the run boundary: identity, session key, timeouts, tools, and terminal outcome. xAI owns the inference response. Keeping those owners separate makes failures easier to name.

## Channel delivery and evidence

Telegram delivery happens after the runtime produces a reply. A valid xAI response therefore does not prove that a Telegram chat is paired, and a paired Telegram chat does not prove that xAI authentication succeeded. Inspect each event source when debugging.

Use xAI's current API reference for provider-specific request details. Avoid hard-coding a model, price, or context-window promise into an architecture explanation because those values can change.

## Frequently asked questions

### Where is the xAI key exposed to the browser?

It is not returned as plaintext. The backend validates and seals it, then supplies the credential to the permitted runtime claim path.

### Does Hermes Agent send requests directly from Telegram to xAI?

No. Telegram is the supported message channel, while the Hermes runtime coordinates the provider-backed run and the eventual channel reply.

## Sources

- [xAI inference API reference](https://docs.x.ai/developers/rest-api-reference/inference) (as of 2026-08-23)
- [xAI model documentation](https://docs.x.ai/developers/models) (as of 2026-08-23)
- [Telegram Bot API](https://core.telegram.org/bots/api) (as of 2026-08-23)

---

- [Full documentation index](/llms.txt)
- [Complete site text](/llms-full.txt)
- [Developer portal](/developers)
- [OpenAPI contract](/openapi.json)
- [MCP server card](/.well-known/mcp.json)