Fix reasoning_details accumulation; filter assistant settings to settings relevant to the selected model

This commit is contained in:
James Babcock committed 2026-11-01 02:20:37 -08:00
1 parent 571c4c84df
commit 41ff1d2733
8 files changed
+170 -37

No files matched your search

+13 -11
View File
@@ -104,16 +104,17 @@ The assistant can call an OpenAI-compatible **Chat Completions** endpoint direct
from the phone. This uses the same conversation history and glasses tool loop as
the built-in cloud providers; it does not require different glasses firmware.
In **Settings > Assistant**, set:
In **Settings > Assistant**:
1. **Compatible API URL** — the explicit API root.
2. **Compatible model ID** — the exact model identifier accepted by the server.
1. **Assistant model** — choose **OpenAI-compatible**. Auto never selects it.
The settings in steps 2–4 appear only while it is selected.
2. **Compatible API URL** — the explicit API root.
3. **Compatible API key** — optional; when set, it is sent only to the configured
compatible endpoint as a Bearer token, not to OpenAI's Whisper service.
4. Optional: choose **Discover models**, then pick a model from the server catalog.
The picker shows tool capability when advertised and hides offline models by
4. **Compatible model ID** — the exact model identifier accepted by the server.
Or choose **Discover models** and pick one from the server catalog. The
picker shows tool capability when advertised and hides offline models by
default. Refresh never changes the saved model by itself.
5. **Assistant model** — choose **OpenAI-compatible**. Auto never selects it.
Provider examples:
@@ -128,11 +129,12 @@ or `/api`. Discovery requests `{API root}/models` only when you choose it; there
is no background discovery or automatic model selection. Manual model entry
remains available for servers without a compatible model-list endpoint.
For the official Venice URL, **Settings > Assistant** also shows **Venice web
search**. It is off by default and visible only for that URL. When enabled,
supported Venice models may search the web for current information; the search
model must support the feature. Open WebUI and other compatible servers never
receive Venice-specific parameters, including after an endpoint switch.
When **OpenAI-compatible** is the assistant model and its URL is the official
Venice URL, **Settings > Assistant** also shows **Venice web search**. It is off
by default and hidden otherwise. When enabled, supported Venice models may
search the web for current information; the search model must support the
feature. Open WebUI and other compatible servers never receive Venice-specific
parameters, including after an endpoint switch.
The same provider can supply Calculator's rewrite fallback and Translate.
Actual tool calling depends on the selected model and server. Use HTTPS where