English | 简体中文 | 日本語 | 한국어 | Deutsch | Français
MyCodex turns Codex + Hermes into a desktop AI workspace for real work: research, sales leads, competitor monitoring, reports, files, and follow-up tasks.
No command-line setup first. No token routing first. Open the app, log in, describe the work, and keep the result, process, and files in one project workspace.
Download · v0.9.7 release notes · What you can do · Screenshots · Feedback
| Platform | Recommended file | Notes |
|---|---|---|
| macOS Apple Silicon | MyCodex-0.9.7-mac-arm64.dmg |
M-series Mac |
| macOS Intel | MyCodex-0.9.7-mac-x64.dmg |
Intel Mac |
| Windows x64 | MyCodex-0.9.0-win-x64.zip |
Older archive package; Windows 0.9.7 is pending |
For manual installation, use the DMG. ZIP files in the release are mainly for the desktop updater.
Most AI agent tools lose users before the first useful task: API keys, shell commands, model settings, scattered files, and no clear place to continue the work.
MyCodex keeps the whole job together:
- the instruction you gave
- the model and project used
- the execution process
- generated files and previews
- follow-up questions
- WeChat handoff when you leave the computer
The value is simple: less setup, less switching, less lost context.
| Workflow | What MyCodex helps with |
|---|---|
| Sales leads | Collect public information, summarize context, and produce follow-up tables |
| Competitor monitoring | Check pages, capture changes, and turn observations into notes |
| Operations reports | Pull scattered material into structured weekly or campaign reports |
| Content research | Gather sources, organize themes, and draft reusable outlines |
| File work | Generate Markdown, CSV, HTML, JSON, images, PDFs, and inspect them in the app |
| Browser tasks | Use a desktop-managed browser bridge for workflows that need real page interaction |
| Mobile handoff | Send tasks and receive result files through WeChat |
This release makes MyCodex better at turning AI work into usable files, and gives local-model users a clearer way to control runtime behavior.
User-visible improvements:
- Local model settings now include runtime parameters, so users can tune
llama-serverlaunch options from the app. - Screenshots, images, PDFs, Office documents, and other generated files are saved as attachments instead of filling the chat with base64 text.
- Artifact cards in the execution process open in the right-side preview panel, so users can inspect results without leaving the conversation.
- Subscription-account and local-model status messages are clearer when model directories or settings fail to load.
- macOS Apple Silicon and Intel installers are refreshed, and update metadata has been published for the desktop updater.
Read the full notes: docs/releases/v0.9.7.md.
Describe the outcome you want, choose a model and project, attach files, or start from a built-in workflow.
See the task, current step, tool actions, and running state instead of waiting in a blank chat box.
The final answer, generated files, attachments, and follow-up context stay in the same conversation.
Preview Markdown, CSV, HTML, JSON, logs, images, PDFs, and other output before opening or downloading.
Connect WeChat to send tasks and receive generated files when you are away from the computer.
Use subscription-account setup when possible, or configure API Key providers when you need more control.
- Founders and managers who want research, summaries, and action lists without opening five tools.
- Operators who repeat reporting, content, community, or campaign workflows.
- Sales teams that need structured lead context and follow-up material.
- Assistants and support teams turning messages, files, and webpages into deliverables.
- Non-technical users who want an agent workspace without starting in a terminal.
MyCodex is early. The best feedback is concrete: what job you tried, where you got stuck, and what output you expected.
MyCodex is currently distributed as packaged installers. Full source is not published in this repository yet.
Test it in a non-critical project first, then move it into workflows where it saves real time.