How to I change the API Host from https://localhost:11434 to something else, such as https://api.example.com or just /api on the same server?
What other environment variables can I set before running npm run build?
Originally created by @coolaj86 on GitHub (Oct 21, 2023).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/8
Originally assigned to: @tjbck on GitHub.
How to I change the API Host from https://localhost:11434 to something else, such as `https://api.example.com` or just `/api` on the same server?
What other environment variables can I set before running `npm run build`?
Currently, our web UI supports the OLLAMA_ENDPOINT variable in the format: OLLAMA_ENDPOINT="http://[insert your ollama URL]", where you must include the port number as well. For instance: OLLAMA_ENDPOINT="http://192.168.4.1:11434/".
If you have any further questions or need assistance, please don't hesitate to ask.
<!-- gh-comment-id:1773918049 -->
@tjbck commented on GitHub (Oct 21, 2023):
Hi,
Currently, our web UI supports the OLLAMA_ENDPOINT variable in the format: OLLAMA_ENDPOINT="http://[insert your ollama URL]", where you must include the port number as well. For instance: OLLAMA_ENDPOINT="http://192.168.4.1:11434/".
If you have any further questions or need assistance, please don't hesitate to ask.
I had running with the ENVs like this, but it didn't work.
SERVER_ENDPOINT='http://localhost:11434' npm run dev
I think that's because I chose SERVER_ENDPOINT, which was only hardcoded.
Do you know how to change the build step to make use of ENVs? It can't be through process.env.* since that doesn't exist at build time... unless it does?
Maybe PageServerLoadcan use process.env.* during the build phase?
<!-- gh-comment-id:1773919838 -->
@coolaj86 commented on GitHub (Oct 21, 2023):
I had running with the ENVs like this, but it didn't work.
```sh
SERVER_ENDPOINT='http://localhost:11434' npm run dev
```
I think that's because I chose `SERVER_ENDPOINT`, which was only hardcoded.
Do you know how to change the build step to make use of ENVs? It can't be through `process.env.*` since that doesn't exist at build time... unless it does?
Maybe `PageServerLoad` _can_ use `process.env.*` during the build phase?
With the latest Pull Request #10 commit, you can use the following command:
OLLAMA_API_ENDPOINT='http://localhost:11434' npm run dev
Let me know if you need further assistance.
<!-- gh-comment-id:1773952195 -->
@tjbck commented on GitHub (Oct 22, 2023):
With the latest Pull Request #10 commit, you can use the following command:
```
OLLAMA_API_ENDPOINT='http://localhost:11434' npm run dev
```
Let me know if you need further assistance.
Model switch: Default bot model changed from OpenAI.gpt-4o-mini to Local.qwen3.5-35b-a3b-fast (local Qwen 3.5 35B)
Free inference (electricity only) vs. paid API
Better quality for French responses
~26s response time on local vLLM (3x RTX 4090)
Async timeout fixes:
aiohttp.ClientTimeout(total=300, sock_read=180) — local models need longer for initial response
Message processing moved to asyncio.create_task — prevents Socket.IO event loop blocking
Typing indicator made fire-and-forget to avoid potential deadlocks
Model visibility fix (all 7 tenants):
Set base_model_id on expert-analyste, redacteur-technique, vision-expert, Local.qwen3.5-35b-a3b-fast
These custom models are now visible to all non-admin users (was broken because base_model_id=None)
Upstream sync: Merged v0.8.8 from upstream (504 commits, including Open Terminal fixes)
Verified
✓ FAQ bot responds using Local.qwen3.5-35b-a3b-fast (26-35s response time)
✓ Replies posted as thread replies in channels
✓ <think> tags stripped from model output
✓ Typing indicator sent without blocking
✓ Bot reconnects automatically after disconnection
✓ Model visible to bot user (26 models total)
Remaining for Phase 3
Multi-tenant deployment (bot per tenant or multi-connection)
RAG integration with knowledge bases
Production monitoring
<!-- gh-comment-id:3990455687 -->
@jsboige commented on GitHub (Mar 3, 2026):
## Phase 2 Update: Local Qwen Model + Async Fixes
### Changes (commit `0b731a7e2`)
- **Model switch**: Default bot model changed from `OpenAI.gpt-4o-mini` to `Local.qwen3.5-35b-a3b-fast` (local Qwen 3.5 35B)
- Free inference (electricity only) vs. paid API
- Better quality for French responses
- ~26s response time on local vLLM (3x RTX 4090)
- **Async timeout fixes**:
- `aiohttp.ClientTimeout(total=300, sock_read=180)` — local models need longer for initial response
- Message processing moved to `asyncio.create_task` — prevents Socket.IO event loop blocking
- Typing indicator made fire-and-forget to avoid potential deadlocks
- **Model visibility fix** (all 7 tenants):
- Set `base_model_id` on `expert-analyste`, `redacteur-technique`, `vision-expert`, `Local.qwen3.5-35b-a3b-fast`
- These custom models are now visible to all non-admin users (was broken because `base_model_id=None`)
- **Upstream sync**: Merged v0.8.8 from upstream (504 commits, including Open Terminal fixes)
### Verified
```
✓ FAQ bot responds using Local.qwen3.5-35b-a3b-fast (26-35s response time)
✓ Replies posted as thread replies in channels
✓ <think> tags stripped from model output
✓ Typing indicator sent without blocking
✓ Bot reconnects automatically after disconnection
✓ Model visible to bot user (26 models total)
```
### Remaining for Phase 3
- Multi-tenant deployment (bot per tenant or multi-connection)
- RAG integration with knowledge bases
- Production monitoring
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Originally created by @coolaj86 on GitHub (Oct 21, 2023).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/8
Originally assigned to: @tjbck on GitHub.
How to I change the API Host from https://localhost:11434 to something else, such as
https://api.example.comor just/apion the same server?What other environment variables can I set before running
npm run build?@tjbck commented on GitHub (Oct 21, 2023):
Hi,
Currently, our web UI supports the OLLAMA_ENDPOINT variable in the format: OLLAMA_ENDPOINT="http://[insert your ollama URL]", where you must include the port number as well. For instance: OLLAMA_ENDPOINT="http://192.168.4.1:11434/".
If you have any further questions or need assistance, please don't hesitate to ask.
@coolaj86 commented on GitHub (Oct 21, 2023):
I had running with the ENVs like this, but it didn't work.
I think that's because I chose
SERVER_ENDPOINT, which was only hardcoded.Do you know how to change the build step to make use of ENVs? It can't be through
process.env.*since that doesn't exist at build time... unless it does?Maybe
PageServerLoadcan useprocess.env.*during the build phase?@tjbck commented on GitHub (Oct 22, 2023):
With the latest Pull Request #10 commit, you can use the following command:
Let me know if you need further assistance.
@ralyodio commented on GitHub (Dec 18, 2023):
How do I change the PORT of the web ui?
@jsboige commented on GitHub (Mar 3, 2026):
Phase 2 Update: Local Qwen Model + Async Fixes
Changes (commit
0b731a7e2)Model switch: Default bot model changed from
OpenAI.gpt-4o-minitoLocal.qwen3.5-35b-a3b-fast(local Qwen 3.5 35B)Async timeout fixes:
aiohttp.ClientTimeout(total=300, sock_read=180)— local models need longer for initial responseasyncio.create_task— prevents Socket.IO event loop blockingModel visibility fix (all 7 tenants):
base_model_idonexpert-analyste,redacteur-technique,vision-expert,Local.qwen3.5-35b-a3b-fastbase_model_id=None)Upstream sync: Merged v0.8.8 from upstream (504 commits, including Open Terminal fixes)
Verified
Remaining for Phase 3