🌎Open WebUI Integration

Open WebUI can connect to Sogni Intelligence through its OpenAI-compatible provider support. This gives self-hosted Open WebUI users access to Sogni's LLMs, streaming chat, vision-capable models, and server-side Sogni media tools.

#Prerequisites

• Open WebUI installed and running.
• A Sogni API key from dashboard.sogni.ai/api-key.
• A Sogni account with Spark or SOGNI token balance.

#Add Sogni As An OpenAI-Compatible Connection

1. Open Open WebUI.
2. Go to Admin Settings.
3. Open Connections.
4. In the OpenAI-compatible connection area, choose OpenAI and add a new external connection.
5. Enter:

Save the connection, then return to chat and open the model selector.

#Model IDs

Use GET https://api.sogni.ai/v1/models for the live model list. Current public model IDs include:

• qwen3.6-35b-a3b-gguf-iq4xs
• qwen3.5-35b-a3b-abliterated-gguf-q4km

#Using Sogni Tools From Open WebUI

Sogni injects media tools by default. If you ask Open WebUI to create an image, video, or music track, /v1/chat/completions can execute the Sogni tool server-side and return Markdown links in the assistant response.

Example prompts:

• Generate a cinematic image of a neon alley in Tokyo during rain.
• Create a five second video of ocean waves crashing on black rocks at sunset.
• Compose a 30 second upbeat synthwave instrumental.

For plain chat or coding-only use, disable Sogni tools if your Open WebUI deployment exposes per-provider request body overrides:

{
  "sogni_tools": false
}

If no such override is available, the default tool list is safe to leave enabled; the model only calls a media tool when the request calls for one.

#Cost Control

Sogni media tools spend tokens. A few rules worth surfacing in your Open WebUI setup:

• Pick the billing token. Send token_type: "spark" (purchased) or token_type: "sogni" (native) via Open WebUI's per-provider request body override if available. Defaults to auto.
• Vendor models need explicit naming. OpenAI GPT Image 2 and ByteDance Seedance 2.0 require Premium Spark and are never picked by the router on your behalf — your prompt or extra-body needs to name them.
• For durable progress / cost approval, use the /v1/chat/runs endpoint with confirm_cost: true and max_estimated_capacity_units directly from your code; Open WebUI itself talks to the synchronous /v1/chat/completions endpoint.

See Billing & Cost Control for the full token-type, gating, and approval flow.

#Vision Notes

Sogni vision input follows the OpenAI content-parts format and requires inline base64 PNG or JPEG data: URIs. Open WebUI image upload behavior can vary by version and deployment. If an image-analysis request fails, confirm the request body sent by Open WebUI uses image_url.url: "data:image/..." rather than a private or remote file URL.

#Troubleshooting

Open WebUI's provider setup may move between versions; see the Open WebUI OpenAI-compatible provider guide for the current UI path.