cd /news/ai-tools/soduwork-custom-model-setup-connect-… · home › topics › ai-tools › article
[ARTICLE · art-170449] src=dev.to ↗ pub= topic=ai-tools verified=true sentiment=· neutral

SoduWork Custom Model Setup: Connect an OpenAI-Compatible Gateway and Handle the 180-Second Stall

A developer connected an OpenAI-compatible gateway (Crazyrouter) to the SoduWork agent workbench v1.13.16 on Windows, registering the model IDs gpt-6-sol and claude-haiku-5-5 by hand after adding a service with an API URL ending in /v1 and an OpenAI-compatible protocol. The setup required manually entering exact catalog model names, since a verified connection alone left the model count at zero, and the developer also reported hitting a 180-second stall during a chat.

by read4 min views1 publishedNov 16, 2026

You install SoduWork, open the chat, and the model picker is empty. What is missing?

A model service. SoduWork is an agent workbench that ships without a language model, so it only works after you connect one. On November 16, 2026, I connected an OpenAI-compatible gateway to SoduWork v1.13.16 on Windows, and both gpt-6-sol and claude-haiku-5-5 appeared in the picker. The short version: add a service under My models, give it an API URL ending in /v1, set the protocol to OpenAI compatible, paste a key, and register each Model ID by hand. This post is for developers who already have an API key and want to switch models inside SoduWork. If your team already provides a preconfigured gateway, skip to the last section on a 180-second stall I hit during a chat.

Item What I used
SoduWork v1.13.16, Windows
Protocol OpenAI compatible
API URL the gateway Base URL, ending in /v1
Credentials a personal API key
Models gpt-6-sol ,claude-haiku-5-5
Checked on November 16, 2026

The official site is soduwork.ai, and installers live on the Releases page of mdiri-ai/SoduWork. At the time of writing, the latest release contained only the Windows installer, SoduWork-1.13.16-Setup.exe; check Releases for other platforms. Forks with the same name exist, so use the official site or that exact repository.

SoduWork does the work and an external model does the thinking, so you need an endpoint first. I used Crazyrouter, where one key reaches models from several vendors. The integration guide shows how the Base URL is written.

Treat the key like a password. A partial key was visible in this screenshot, and I masked it before publishing.

My list already had a company gateway, ByteDance Volcano Ark, DeepSeek, Alibaba Cloud Bailian, Zhipu GLM, and Kimi. They use the same screen, so connecting a vendor directly follows the same steps.

Field Value
Service name Crazyrouter (any recognizable name)
API URL copy the line below
API protocol OpenAI compatible
No key required leave unchecked
https://api.crazyrouter.com/v1

The trailing /v1 is the part that gets forgotten. Before a key is pasted, Connection status reads Verification failed, which is just the initial state.

Paste the key into Personal API key under Access credentials and click Save and test connection. When the status becomes Connection verified, click Save.

A verified connection means the service is reachable, nothing more. Model count was still 0. Where do the IDs come from?

I used the Model Square on the Crazyrouter site, which listed 173 models with vendor and type filters. The model catalog works for checking names as well. In SoduWork:

Field Value
Model ID the exact catalog name, such as gpt-6-sol
Display name same as the ID, or any label you prefer
Model purpose Chat model
Reasoning mode left on Auto detect
Model capabilities image input, tool call, and streaming were all Supported for me

A single wrong character in the Model ID makes calls fail, so copy the name instead of retyping it.

Save, then repeat for claude-haiku-5-5. One gateway now exposes models from two vendors.

Click New chat, then the model name at the bottom right of the input box to open Choose a model. Both models show their source as Crazyrouter · Personal quota.

The Reasoning mode at the bottom of that window said Unsupported for gpt-6-sol; I left it alone. The company gateway's glm-5.3-flash appeared in the same list, so models from different sources can be mixed.

During a chat, a task stopped partway with a notice that 180 seconds had passed. From what I could check, the limit is a fixed threshold built into the app. My models has no matching option, and changing the gateway does not move it.

What I would do about it:

Symptom Check first
Verification failed URL ends in /v1 , protocol is OpenAI compatible, key pasted in full, account has balance
Model missing from the picker Model ID matches the catalog exactly, Save was clicked
A chat call fails Re-copy the Model ID, confirm personal quota remains
Another vendor needed Add a second service in My models

This covers only what I verified: the setup flow and model selection on SoduWork v1.13.16 for Windows. I did not benchmark how each model handles long tasks inside SoduWork. If you have connected several models, has the 180-second limit ever stopped one of your sessions?

References: SoduWork official site and SoduWork GitHub Releases.

── more in #ai-tools 4 stories · sorted by recency
── more on @soduwork 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/soduwork-custom-mode…] indexed:0 read:4min 2026-11-16 · —