---
title: Choosing Models for Hermes Agent
description: "Choose a model policy for Hermes Agent by separating task requirements, provider credentials, tools, fallbacks, and operational evidence."
canonical: "https://deploy-hermes.com/technical/choosing-models-for-hermes-agent"
last-updated: "2026-08-23"
---

# Choosing Models for Hermes Agent

> Choose a model policy for Hermes Agent by separating task requirements, provider credentials, tools, fallbacks, and operational evidence.

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

Model choice in Hermes Agent is a policy decision inside a larger runtime. Select a provider and model path that fit the agent's job, then verify the behavior with the same tools, channel, and controls the production run will use.

## Translate the task into requirements

List the task's input shape, reasoning burden, tool needs, latency tolerance, privacy boundary, and failure cost. An agent that summarizes an internal note should not inherit the same model policy as one that coordinates a multi-step operational task.

Include the runtime's own work in the test: prompt composition, tool descriptions, memory or session context, approval pauses, and the channel's message limits. These can change what “good model output” means.

## Keep provider configuration explicit

Connect only the credentials the workspace needs and order providers intentionally. A fallback chain is a behavior change: it can affect output style, tool support, latency, and where a request is processed. Document the reason for each fallback and how operators identify it in a run.

Use provider documentation for current model catalogs and API constraints. Do not make an exact model identifier, price, or context size the definition of an agent's architecture.

## Prove the selected policy

Start with a low-risk test and record the provider, model, tool outcomes, duration, and final state. Repeat with an ambiguous request and a provider failure so you know whether the agent stops, asks for help, or falls back.

Revisit the policy when the task, provider catalog, credential owner, or safety requirements change. A model policy is executable configuration, not a one-time marketing preference.

## Frequently asked questions

### Can Hermes Agent use more than one provider?

Yes. Supported provider credentials can participate in the runtime's configured provider policy, including a deliberate fallback chain.

### Should I configure a fallback on day one?

Only if you can test and explain it. A single well-observed provider is often easier to operate than an untested chain with unclear behavior.

## Sources

- [xAI model documentation](https://docs.x.ai/developers/models) (as of 2026-08-23)
- [OpenAI models documentation](https://developers.openai.com/api/docs/models) (as of 2026-08-23)
- [Anthropic model list](https://platform.claude.com/docs/en/api/models/list) (as of 2026-08-23)
- [Google Gemini models](https://ai.google.dev/gemini-api/docs/models) (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)