Where it is today
The current release is v0.1.2, in private daily use on its author’s Linux machines. It runs in a terminal. The repository is private, and there is no public release date yet.
You can ask a configured model, read streaming answers, search your history as you type, and copy commands directly from results. Exact repeats return saved answers without another API call.
Current limits:
- Hotkey integration is next. Automated compositor setup and hardware verification are planned for v0.2.0.
- Model choice is in configuration. There is no in-app model picker yet.
- Semantic caching is opt-in. Related questions aren’t automatically served from memory by default. Semantic search is available separately with a local embedding model.
- History stays on one machine. Sync isn’t available yet.
- No API spending limit. Budget controls are planned.
- Some repairs are limited. Repairing reopened answers and moving a mismatched question between privacy stores aren’t supported yet.
Phase 1: daily driver shipped
Ask a question, get an answer, and find it again when you need it.
- Streaming answers from one configured model
- Formatted Markdown and highlighted code
- Local history and exact-match caching
- Keyword search as you type
- Copy a code line, code blocks, or the whole answer
- Copy directly from search results
- Private requests restricted to eligible providers
- A background service that starts when needed
Hotkey configuration is documented, but automated setup and hardware verification belong to the v0.2.0 release.
Phase 2: the memory planned
Make the answer you return to more useful over time.
Semantic search, opt-in semantic caching, feedback on live answers, and soft delete with undo already work. The remaining milestone includes:
- Background distillation of responses into better titles and saved notes
- Controls to freeze a title or answer you want to keep
- Better handling of incomplete answers and incorrectly grouped questions
- Re-ask statistics to show what you keep coming back for
- Permanent deletion controls
Automatic semantic reuse remains opt-in because matching quality hasn’t met the bar for enabling it by default.
Phase 3: parallel and portable planned
Get another model’s perspective without starting over, and take your history to your other machines.
- Ask several models at once: read the fast answer while others work
- Switch answers, stop a model, or add one while a question is running
- Sync eligible history between machines
- Keep private and local-only records outside sync
- Connect a lightweight client to a daemon on another machine
Phase 4: later planned
Longer-term directions, still unscheduled:
- Optional context from notes, Obsidian, and MCP sources
- Export answers to other tools
- Combine and compare responses
- API budgets and spending limits
- A phone client
- Model selection tailored to the question
- Suggestions for separating questions that were grouped incorrectly
- Public release, including licensing and packaging
Releases
Phases describe capabilities. Releases package the work that’s ready to use.
| Release | What it brings | Status |
|---|---|---|
| v0.1.0–v0.1.2 | The terminal app, installation, provider setup, and setup fixes. | Released for private use |
| v0.2.0 | Instant activation: automated compositor setup, verified hotkey toggle, Omarchy menubar integration, and theming. | Planned |
| v0.3.0 | A native Wayland window, giving Myellm its own surface. | Future |
Planned scope can change. Public availability is a separate milestone with no date attached. See the changelog for released changes.
What guides the work
Build the smallest thing worth using every day, then improve it through repeated use. Keep typing and local recall responsive. Preserve answer history. Make mistakes visible and recoverable.
A useful answer should be easier to get back to each time you need it.