What feature would you like to see?
I would love to see a more complete power-user / local-AI mode in Codexia. I searched the current Codexia issues first and did not find an existing issue covering the main requests below.
1. French localization / i18n
Please add a real French UI (fr-FR) and ideally make the localization system extensible for other languages too.
Suggested UX:
- Auto-detect the OS language on first launch.
- Manual language selector in Settings.
- Keep English as fallback when a translation key is missing.
- Translate menus, settings, permission prompts, errors, updater text, model/provider UI, plugins/skills UI, etc.
This would make Codexia much easier to use for people who are not comfortable with English.
2. OpenAI Plugin Directory / Codex plugin integration
OpenAI now has a Plugin Directory for ChatGPT and Codex. Plugins can package skills, apps and app templates. Codexia already has MCP/skills marketplace functionality, so it would be very useful if Codexia could integrate with the official Codex/OpenAI plugin ecosystem when a supported API or format is available.
Possible approaches:
- Browse/import/sync compatible official Codex plugins from Codexia.
- Import plugin/skill/MCP manifests or compatible configurations.
- Show which capabilities a plugin needs before enabling it.
- Preserve the original authentication, workspace and permission rules instead of bypassing them.
- Clearly distinguish official OpenAI/Codex plugins from community/local plugins.
Official OpenAI reference:
https://help.openai.com/en/articles/20001256-plugins-in-codex/
3. Optional Computer Use / real desktop control
Full Access is useful for filesystem/shell access, but it is not the same thing as full GUI/desktop control.
When the selected provider/model supports it, it would be great to expose an optional Computer Use capability for screenshots, mouse, keyboard, browser/UI interaction, etc.
Suggested safeguards:
- Explicit opt-in per session or profile.
- Separate permissions for shell, filesystem, network, MCP/plugins and Computer Use.
- Visible activity/audit log.
- Confirmation for sensitive actions.
- Ability to disable GUI control while keeping shell/file access.
For local models, Codexia could allow compatible MCP/Playwright/desktop-automation tools rather than pretending every local model natively supports OpenAI Computer Use.
OpenAI documents Computer Use as a Codex capability and as a Responses API tool on supported models:
https://help.openai.com/en/articles/11369540-using-codex-with-your-chatgpt-plan
4. First-class LM Studio / OpenAI-compatible local provider UX
The recent provider work on master is already a big improvement. I would love to see it expanded into a polished local-provider experience:
- Add/edit custom provider profiles with
Base URL, optional API key and model ID.
- Support local and LAN endpoints (for example, LM Studio running on another PC).
- Auto-discover
/v1/models.
- One-click Test connection.
- Test
/v1/responses compatibility, not only the model list.
- Manual refresh of discovered models.
- Do not require a fake API key when the local server has authentication disabled.
- Show useful model capability badges when detectable: vision, tools/function calling, reasoning, embeddings, context size, etc.
- Friendly error messages explaining whether the failure is DNS/network, model discovery, authentication or Responses API compatibility.
I recently tested a remote LM Studio server with Codexia on Windows. /v1/models and /v1/responses worked correctly, but v0.46.0 still needed a localhost workaround because of the older custom-provider setup. The new user-managed provider work on master appears to address the base-URL part, so this request is mainly for the remaining UX/diagnostics/polish.
5. Built-in local model manager + hardware-aware recommendations
It would be extremely useful if Codexia could optionally detect the user's hardware and help choose a local model instead of requiring the user to already understand parameter counts, quantization and VRAM requirements.
Possible workflow:
- Detect OS/architecture, CPU, system RAM, GPU(s), VRAM and available memory.
- Ask what the user wants to do: coding, general chat, vision, tool use, fast/lightweight, best quality, etc.
- Recommend a few models and quantizations that realistically fit the machine.
- Explain why each model is recommended and the expected RAM/VRAM/context trade-offs.
- Estimate resources before loading when the backend supports it.
- Let the user confirm before any model download or large disk usage.
LM Studio now exposes a native model-management REST API, including model download, download status, load and unload endpoints, so Codexia could optionally integrate with it instead of reimplementing a model runtime:
https://lmstudio.ai/docs/developer/rest
For example, LM Studio documents:
POST /api/v1/models/downloadGET /api/v1/models/download/status/:job_idPOST /api/v1/models/load- model load/resource estimation through its tooling
It could later support other backends such as Ollama through the same abstraction.
6. Update UX / settings
I initially wanted to request automatic update detection and one-click install/restart. I noticed that master already contains recent updater work, so I do not want to duplicate that implementation.
What would still be useful is a clear Settings section such as:
Check for updates automaticallyNotify me when an update is readyStable / prerelease channel (if supported)Download automatically, install only after confirmation- Show current version and latest version
- One-click install + restart
7. More granular permission profiles
In addition to the current broad access modes, it would be useful to have advanced toggles/profiles such as:
- Filesystem read/write
- Shell/terminal execution
- Network access
- MCP/plugins
- Clipboard
- Browser automation
- Computer Use / mouse + keyboard + screenshots
A user who explicitly wants the most capable local setup could choose a clearly labeled Maximum local access profile, while other users can keep safer defaults.
8. One-click diagnostics for providers/models
A small diagnostics page would make local-AI setup much easier:
- Provider reachable: yes/no
/v1/models: pass/fail/v1/responses: pass/fail- Selected model found: yes/no
- Tool/function-calling support test
- Vision support test when applicable
- Authentication state
- Latency / simple token-speed test
- Copy/export diagnostics for GitHub bug reports
Additional information
I checked the issue tracker before posting. I did find the existing Windows PowerShell request (#43), so I intentionally did not duplicate terminal-shell selection here.
I also noticed two recent improvements on master that overlap partially with this proposal:
- user-managed model providers / custom Base URLs
- updater improvements
So those parts above should be read as extensions/polish, not requests to redo work that has already landed.
This is intentionally an umbrella request because the features work together as a "complete local/power-user experience". If the project prefers one feature per issue, I am happy for this to be split into smaller issues.
What feature would you like to see?
I would love to see a more complete power-user / local-AI mode in Codexia. I searched the current Codexia issues first and did not find an existing issue covering the main requests below.
1. French localization / i18n
Please add a real French UI (
fr-FR) and ideally make the localization system extensible for other languages too.Suggested UX:
This would make Codexia much easier to use for people who are not comfortable with English.
2. OpenAI Plugin Directory / Codex plugin integration
OpenAI now has a Plugin Directory for ChatGPT and Codex. Plugins can package skills, apps and app templates. Codexia already has MCP/skills marketplace functionality, so it would be very useful if Codexia could integrate with the official Codex/OpenAI plugin ecosystem when a supported API or format is available.
Possible approaches:
Official OpenAI reference:
https://help.openai.com/en/articles/20001256-plugins-in-codex/
3. Optional Computer Use / real desktop control
Full Accessis useful for filesystem/shell access, but it is not the same thing as full GUI/desktop control.When the selected provider/model supports it, it would be great to expose an optional Computer Use capability for screenshots, mouse, keyboard, browser/UI interaction, etc.
Suggested safeguards:
For local models, Codexia could allow compatible MCP/Playwright/desktop-automation tools rather than pretending every local model natively supports OpenAI Computer Use.
OpenAI documents Computer Use as a Codex capability and as a Responses API tool on supported models:
https://help.openai.com/en/articles/11369540-using-codex-with-your-chatgpt-plan
4. First-class LM Studio / OpenAI-compatible local provider UX
The recent provider work on
masteris already a big improvement. I would love to see it expanded into a polished local-provider experience:Base URL, optional API key and model ID./v1/models./v1/responsescompatibility, not only the model list.I recently tested a remote LM Studio server with Codexia on Windows.
/v1/modelsand/v1/responsesworked correctly, but v0.46.0 still needed a localhost workaround because of the older custom-provider setup. The new user-managed provider work onmasterappears to address the base-URL part, so this request is mainly for the remaining UX/diagnostics/polish.5. Built-in local model manager + hardware-aware recommendations
It would be extremely useful if Codexia could optionally detect the user's hardware and help choose a local model instead of requiring the user to already understand parameter counts, quantization and VRAM requirements.
Possible workflow:
LM Studio now exposes a native model-management REST API, including model download, download status, load and unload endpoints, so Codexia could optionally integrate with it instead of reimplementing a model runtime:
https://lmstudio.ai/docs/developer/rest
For example, LM Studio documents:
POST /api/v1/models/downloadGET /api/v1/models/download/status/:job_idPOST /api/v1/models/loadIt could later support other backends such as Ollama through the same abstraction.
6. Update UX / settings
I initially wanted to request automatic update detection and one-click install/restart. I noticed that
masteralready contains recent updater work, so I do not want to duplicate that implementation.What would still be useful is a clear Settings section such as:
Check for updates automaticallyNotify me when an update is readyStable / prerelease channel(if supported)Download automatically, install only after confirmation7. More granular permission profiles
In addition to the current broad access modes, it would be useful to have advanced toggles/profiles such as:
A user who explicitly wants the most capable local setup could choose a clearly labeled Maximum local access profile, while other users can keep safer defaults.
8. One-click diagnostics for providers/models
A small diagnostics page would make local-AI setup much easier:
/v1/models: pass/fail/v1/responses: pass/failAdditional information
I checked the issue tracker before posting. I did find the existing Windows PowerShell request (#43), so I intentionally did not duplicate terminal-shell selection here.
I also noticed two recent improvements on
masterthat overlap partially with this proposal:So those parts above should be read as extensions/polish, not requests to redo work that has already landed.
This is intentionally an umbrella request because the features work together as a "complete local/power-user experience". If the project prefers one feature per issue, I am happy for this to be split into smaller issues.