I've deployed Sourcebot v4.17.2 (via the official container image). I'm using it with an on-premise Gitea instance. Repo indexed worked flawlessly. For the "ask" feature I've tried multiple models via ollama:
"models": [
{
"provider": "openai-compatible",
"baseUrl": "http://...",
"model": "qwen3-coder:30b",
"displayName": "qwen3-coder:30b"
}
Irregardless of which model I try, answers always stop after the first output - sometimes with a plan about what to do (without continuing from there), at other times just with a description of the tool function to use (e.g. <function=list_repos> </tool_call>). Context seems lost after that.
I've set SOURCEBOT_LOG_LEVEL to debug, but all I can see are messages regarding indexing, or something like
[web] | 2026-05-22T11:25:56.031Z debug: [tool] list_repos {"page":1,"perPage":30,"sort":"name","direction":"asc"}
(this was after it answered with <function=list_repos> </tool_call>).
Any idea what could be wrong? I'm happy to try other models, but I'm limited to models up to ~48GB of VRAM.
I've deployed Sourcebot v4.17.2 (via the official container image). I'm using it with an on-premise Gitea instance. Repo indexed worked flawlessly. For the "ask" feature I've tried multiple models via ollama:
Irregardless of which model I try, answers always stop after the first output - sometimes with a plan about what to do (without continuing from there), at other times just with a description of the tool function to use (e.g.
<function=list_repos> </tool_call>). Context seems lost after that.I've set
SOURCEBOT_LOG_LEVELtodebug, but all I can see are messages regarding indexing, or something like(this was after it answered with
<function=list_repos> </tool_call>).Any idea what could be wrong? I'm happy to try other models, but I'm limited to models up to ~48GB of VRAM.