Description
After migrating from Copilot v3.x to v4, BYOK custom provider models configured under Settings > BYOK and enabled in Settings > Agents > OpenCode do not appear in either the Copilot Chat pane model selector or the new Copilot Agent pane model selector.
Steps to Reproduce
- Migrate from Copilot v3.x to v4.
- Go to Settings > BYOK and add a custom provider (e.g.
oMLX with its API key and Base URL). Check the box for the desired model(s) (e.g. gemma-4-e4b-it-8bit).
- Repeat for another provider (e.g.
OpenRouter), adding a model such as Qwen3 235B A22B Instruct 2507.
- Go to Settings > Agents > OpenCode and toggle on all BYOK models — they appear in the list and the toggle saves correctly.
- Open the Copilot Chat pane and click the model selector.
- Open the Copilot Agent pane and click the model selector.
Expected Behaviour
The BYOK models enabled in the OpenCode agent configuration (gemma-4-e4b-it-8bit under oMLX, Qwen3 235B A22B Instruct 2507 under OpenRouter) should appear as selectable options in both the Chat and Agent model selectors.
Actual Behaviour
- Copilot Agent pane model selector only shows built-in Claude models (Default/Sonnet 4.6, Opus 4.8, Haiku 4.5). No BYOK models listed.
- Copilot Chat pane model selector shows various built-in/known provider models but none of the custom BYOK provider models.
Screenshots
(Attach the three screenshots shared above)
OpenCode agent settings — BYOK models toggled on:
- oMLX (BYOK):
gemma-4-e4b-it-8bit — ON
- OpenRouter (BYOK):
Qwen3 235B A22B Instruct 2507 — ON
Copilot Agent pane model selector: Only Default (Sonnet 4.6), Opus (4.8), Haiku (4.5) — no BYOK models.
Copilot Chat pane model selector: copilot-plus-flash plus various GPT/Gemini/Claude models — no oMLX or OpenRouter BYOK models.
Environment
- Obsidian Copilot version: v4.x (migrated from v3.x)
- Local model provider: oMLX (custom OpenAI-compatible endpoint)
- Additional BYOK provider: OpenRouter
- OS: macOS
Notes
The OpenCode agent settings correctly show and persist the toggle state for BYOK models — the disconnect appears to be between the agent model configuration and the model selectors surfaced in the Chat/Agent UI panes. This worked as expected in v3.x.
Description
After migrating from Copilot v3.x to v4, BYOK custom provider models configured under Settings > BYOK and enabled in Settings > Agents > OpenCode do not appear in either the Copilot Chat pane model selector or the new Copilot Agent pane model selector.
Steps to Reproduce
oMLXwith its API key and Base URL). Check the box for the desired model(s) (e.g.gemma-4-e4b-it-8bit).OpenRouter), adding a model such asQwen3 235B A22B Instruct 2507.Expected Behaviour
The BYOK models enabled in the OpenCode agent configuration (
gemma-4-e4b-it-8bitunderoMLX,Qwen3 235B A22B Instruct 2507underOpenRouter) should appear as selectable options in both the Chat and Agent model selectors.Actual Behaviour
Screenshots
(Attach the three screenshots shared above)
OpenCode agent settings — BYOK models toggled on:
gemma-4-e4b-it-8bit— ONQwen3 235B A22B Instruct 2507— ONCopilot Agent pane model selector: Only Default (Sonnet 4.6), Opus (4.8), Haiku (4.5) — no BYOK models.
Copilot Chat pane model selector: copilot-plus-flash plus various GPT/Gemini/Claude models — no oMLX or OpenRouter BYOK models.
Environment
Notes
The OpenCode agent settings correctly show and persist the toggle state for BYOK models — the disconnect appears to be between the agent model configuration and the model selectors surfaced in the Chat/Agent UI panes. This worked as expected in v3.x.