Summary
Google Gemini provider requests return 404 errors and the model list cannot be fetched because Marinara Engine appends /v1/ instead of /v1beta/ to the endpoint.
Expected behavior
The Google Gemini provider should use the /v1beta/ endpoint so Gemini model requests and model list fetching succeed.
Actual behavior
Marinara Engine uses the /v1/ endpoint, causing 404 errors when calling Gemini models and preventing the model list from being fetched.
Steps to reproduce
- Install or run Marinara Engine 2.4.2 (dd77689) through Termux on Android 16.
- Configure the Google Gemini provider.
- Attempt to fetch the model list or call a Gemini model.
- Observe a 404 response caused by the request using
/v1/ instead of /v1beta/.
Environment
- Marinara Engine version: 2.4.2 (dd77689)
- Install type: termux
- OS + version: Android 16
- Browser or app shell: Not provided
Logs, screenshots, or video
No logs or attachments provided. Gemini API method reference: https://ai.google.dev/api/all-methods
Additional context
The reporter initially referred to SillyTavern but clarified that the affected application is Marinara Engine.
Discord thread: Bug report - Gemini provider has wrong endpoint
Discord thread link: https://discord.com/channels/1417099416812392641/1536304459217371146
Requested by: Marinara (marinara_spaghetti, Discord ID 169561011928432653)
Discord thread transcript (6 messages)
[2026-08-10T09:25:26.234Z] Fussel: ME version: 2.4.2 (dd77689)
Installation method: Termux (Android 16)
When I try to call a Gemini model over the Google Gemini provider in Marinara Engine it results in a 404 error.
The reason it does that (and does not let me fetch the model list) is because the endpoint ME appends /v1/ when it should be with /v1beta/
See:
https://ai.google.dev/api/all-methods
All methods | Gemini API | Google AI for Developers
[Discord message: https://discord.com/channels/1417099416812392641/1536304459217371146/1536304459217371146]
[2026-08-10T09:31:58.268Z] Marinara: And this is ME’s problem how?
[Discord message: https://discord.com/channels/1417099416812392641/1536304459217371146/1536306103527153734]
[2026-08-10T09:32:06.405Z] Marinara: You should raise it on ST forum.
[Discord message: https://discord.com/channels/1417099416812392641/1536304459217371146/1536306137656336495]
[2026-08-10T09:33:14.719Z] Fussel: Sorry, for some reason I wrote SillyTavern and meant Marinara Engine 😅
[Discord message: https://discord.com/channels/1417099416812392641/1536304459217371146/1536306424186015846]
[2026-08-10T09:33:31.060Z] Fussel: That's on me
[Discord message: https://discord.com/channels/1417099416812392641/1536304459217371146/1536306492725002302]
[2026-08-10T09:33:37.716Z] Marinara: (no text content)
[Discord message: https://discord.com/channels/1417099416812392641/1536304459217371146/1536306520642420876]
Template check
Please uncheck (untick) the box below before submitting so we know you read the template. It is intentionally pre-checked:
Summary
Google Gemini provider requests return 404 errors and the model list cannot be fetched because Marinara Engine appends
/v1/instead of/v1beta/to the endpoint.Expected behavior
The Google Gemini provider should use the
/v1beta/endpoint so Gemini model requests and model list fetching succeed.Actual behavior
Marinara Engine uses the
/v1/endpoint, causing 404 errors when calling Gemini models and preventing the model list from being fetched.Steps to reproduce
/v1/instead of/v1beta/.Environment
Logs, screenshots, or video
No logs or attachments provided. Gemini API method reference: https://ai.google.dev/api/all-methods
Additional context
The reporter initially referred to SillyTavern but clarified that the affected application is Marinara Engine.
Discord thread: Bug report - Gemini provider has wrong endpoint
Discord thread link: https://discord.com/channels/1417099416812392641/1536304459217371146
Requested by: Marinara (marinara_spaghetti, Discord ID 169561011928432653)
Discord thread transcript (6 messages)
Template check
Please uncheck (untick) the box below before submitting so we know you read the template. It is intentionally pre-checked: