Hosting

Choose a model

Martian Security inference stack

Qwen 3.8, GLM 5.3 Flash and DeepSeek V4.1 Flash 1017 abliterated models are hosted in the Martian Security inference stack and served through our OpenAI-compatible APIs.

Existing configurations

Setup	Provider
Fresh template	DeepSeek V3.2 through Bedrock, using the EC2 role
Hosted profile documented on 1 Nov 2026	GPT-6.1 Sol · medium · ChatGPT/Codex OAuth

Change the provider with hermes model or Models → Change. Start a new chat afterward; existing sessions can retain their earlier model.

Hermes Models page showing openai-codex and gpt-6.1-sol as the main model — Actual saved settings, 30 October 2026. History includes earlier models. Open full size

Bedrock IAM

Give the EC2 role access to the configured model in the chosen region. Verify an inference turn with a tool call. Bedrock setup

ChatGPT / Codex OAuth

Back up first. As hermes:

hermes auth add openai-codex --no-browser
hermes config set model.default gpt-6.1-sol
hermes config set agent.reasoning_effort medium

The account owner completes sign-in. Hermes stores its own credentials in HERMES_HOME; do not copy the operator’s Codex credentials or paste tokens into chat.

Device login fails, or the reasoning setting warns

Use hermes auth add openai-codex --browser --no-browser for browser OAuth. Forward callback port 1455 → 1455 through SSM as well as the dashboard port, then open the displayed authorization link.

The pinned config checker warns about agent.reasoning_effort while saving it. Confirm Medium in a new chat and test an actual inference turn. A listed model does not prove account access or remaining quota.

Promptfoo, PyRIT and garak need their own provider settings. Hermes OAuth does not configure them. Hermes provider reference