Roadmap

Build toward coordinated AI work.

Explore the planned AI team workflow for research, proposal review and controlled execution. Meeting and reading tools already implemented in development builds support the records and evidence behind your work.

The order and proposed scope may change as testing continues. Completion dates are not set.

Planned experience

Your AI team

Research, design and review from sources you choose are still in development.

See the team scope
Built in development

Recording and knowledge

Recording, transcripts, summaries and search are implemented in the development build.

Explore the current build
Built in development

Reading and research

Side-by-side translation, PDF page translation, source-linked notes and document questions are implemented.

Explore the reading workspace
Research only

Visual planning

A planning board will be considered only if trials show it helps more than lists and search.

See the research direction

AI team

Planned first AI team release

The proposed workflow uses recordings and other sources you choose. Discuss with AI members, adjust their roles and skills, then research, design and execute together. These capabilities are still planned.

Planned

First-release priorities

10 items

Team discussion, proposal review and controlled execution.

  • Explore questions through research rounds
  • Discuss directly with AI members, with names and roles visible
  • Use recommendation cards: adopt, adopt provisionally, defer
  • Review a design proposal and accept a specific version
  • Break a work plan into clear tasks
  • Run tasks, pause, request all work to stop, or take over
  • Use research-to-decision and material-to-document templates
  • Use Codex with your ChatGPT plan, or Claude Code
  • Save and reuse members, teams and cases
  • Validate the workflow with one representative coding task
Planned

Selected-case support

4 items

Selected cases; separate validation.

  • Control a browser for a selected case
  • Use the desktop for a separately selected case
  • Connect read-only to an existing Codex or Claude Code session
  • An optional AI coordinator proposes assignments for each round; results are reviewed separately
Planned

Later considerations

4 items

Outside the first release.

  • Run scheduled or event-triggered routine work
  • Observe ChatGPT Desktop or Claude Chat sessions
  • Run comparisons between different versions of a case
  • Export saved member and team settings, and case definitions, in native formats

This is proposed scope, not released functionality. Integrations still need validation and the release may differ. Choosing an external AI service sends relevant source material to that provider.

See the agent team screens

Sources and knowledge

Development order for sources and workflows

Continue improving recording and storage, make answers easier to inspect, retain project context and reuse workflows. The reading workspace is implemented; AI teamwork and further integration are planned alongside these stages.

  1. Stage 0: Recording and storage Improve recording, storage and recovery before release. Current focus Stage details

    Recording, transcripts, notes, summaries and the note map working end to end, with truthful cancellation, search, deletion and recovery.

    Moves on wheneach release candidate passes its reliability checks on the platform it ships for.

  2. Stage 1: Selected sources and evidence Choose sources and inspect cited answers and work records. Planned Stage details

    Choose sources for a question, inspect cited passages and review the work records available.

    Moves on whensource choices hold end to end, and cancelled work can never publish.

  3. Stage 2: Project continuity Resume projects with current decisions and unfinished work. Planned Stage details

    Pick a project back up, keep commitments honest, and review what the knowledge says.

    Moves on whenpeople recover the current decision and open commitments without mixing history, proposals and uncertainty.

  4. Stage 3: Workflow reuse and intake Reuse workflows and choose which material to import. Planned Stage details

    Repeat good work safely, and bring in outside material on purpose.

    Moves on whenevery workflow, preference and intake path has clear limits, a preview and a way to revoke it.

  5. Stage 4: Evaluate a planning board Test a planning board against lists and search. Research only Stage details

    A planning board, only if research shows it beats lists and search for a real task.

    • Planning board

    Moves on whentrials show a clear improvement over the simpler view.

AI teamwork and reading integration

Build on the implemented reading workspace to connect selected sources with future AI teamwork. Team integration, recaps, bilingual meetings and new intake paths are planned in parallel, with separate scope and validation.

These stages cover broader source and knowledge workflows. The current reader already supports translation, notes and cited questions. Each new workflow still needs its own validation before release.

Built in Studio development

Five screens already implemented in Studio.

These five screens show the recording features implemented in the current development build, with English and Traditional Chinese interfaces. The reading workspace is also implemented; AI teamwork remains planned.

Live transcript with your notes alongside, so you can stay in the conversation

Records microphone and system audio together, with live level meters, pause and resume. The transcript appears as people speak, and your notes sit on the same screen. The backend saves committed work independently of the window; interrupted recordings can still have gaps or unavailable audio, and installed recovery testing remains in progress.

The recording workspace: notes and the live transcript in view together, the summary and recall panes a tab away, and the controls always within reach.

Every meeting in one place, with progress and action items at a glance

Every meeting's recording, transcript, summary and notes live here and can be searched together, and an action items tab gathers what each meeting left open. Badges on the cards say plainly which meetings have a summary and which have only a transcript. Imported audio files go through the same pipeline.

The library: a tab for meetings and a tab for action items, search across both, and a badge on any card whose recording is not yet complete.

Decisions and action items in a summary that says where it was made

The transcript and the AI summary side by side. Write your own summary templates, such as executive briefings, client discovery or technical standups, to pull out key points and concrete action items. Every summary records where it was generated, on your device or at an endpoint you configured, and can be exported to Markdown, PDF or HTML at any time.

A recorded meeting: the transcript, the summary and its action items, with the template, the summary language, and the attribution line that records where the text was generated.

Ask your past meetings and check the sources used

Ask questions about one recorded meeting or across your meetings. Answers use meeting records as their basis. Open a numbered source attached to an answer to see the cited transcript passage and cue the recording to that time. Check important answers against the original record.

Ask: a scope switch between all meetings and this meeting, suggested questions, and the sources attached to the answer.

Choose your own models, trading off privacy, cost and speed

You decide which AI models live on your computer: each download starts only when you ask, can resume after an interruption, and is verified against a pinned checksum. Each model is labelled with where it runs, whether it is installed and whether it needs an API key. The default models were chosen for English and Traditional Chinese.

The model catalog, inside settings: each row tagged by where it runs and whether it is installed, with one download in progress. Model names are omitted here, not in the application.

Read how it is built

Built in the development build

Reading and research are already here.

Import a PDF, Word (.docx), Markdown or text file. Keep the original, translation, notes and questions in one reading workspace.

Implemented in development builds

  • Read in two languages

    Switch between Original, Translation and Side by side. Translate source content into Traditional Chinese for Taiwan while keeping the original available to check.

    Read the reading guide

  • Translate a page or document

    PDF translations appear over the source page layout. Translate the current page or the document, stop when needed and continue from saved progress. Word, Markdown and text use a flowing reading view.

    Read the reading guide

  • Follow citations back to the source

    Follow supported numbered references in the document, or open the sources cited in an Ask answer. Highlight passages and quote them into Markdown notes with their source links.

    Read the reading guide

  • Study and ask questions

    Generate Study notes with section key points and passage links. Ask a question about this document or all documents, then open the cited page or section to check the answer.

    Read the reading guide

Implemented in development builds; public installers are not available yet. Translation and Study notes need a built-in model; Ask uses your selected provider. PDF layout and translation quality depend on the source. Scanned PDFs support reading and notes, but translation, highlights and Ask need extractable text. AI teamwork remains planned.

Try the reading demo

In this release

Highlights of version 0.1.0

Recording highlights in the current development build. Reading features are covered above. Public installers are not available yet.

Version v0.1.0

  1. Ask

    Answers you can check.

    Open a numbered source attached to an Ask answer to revisit the cited transcript passage, with the recording cued to that time.

    Explore this capability — Ask
  2. Windows setup

    Setup that fits the computer.

    Studio’s Windows setup checks hardware, runtime and models before enabling live summaries. The 8 GB video-memory and 16 GB system-memory thresholds are provisional. Without live summaries, supported configurations can record, transcribe and summarize afterwards.

    Explore this capability — Windows setup
  3. Usage & exports

    See your activity. Take it with you.

    Review recording activity by date, then save a CSV or an English or Traditional Chinese PDF report with charts and optional daily details.

    Explore this capability — Usage & exports
  4. Audio playback

    Return to the moment that matters.

    Keep playback position when switching transcript and map layouts. Preview a position while scrubbing before seeking, with ten-second controls, playback speed and transcript-follow controls.

    Explore this capability — Audio playback
  5. Meeting reading

    Switch between Summary and Map.

    Saved meetings have separate Summary and Map views that preserve the summary editor when switching. Transcripts add reading density, visible timestamps and search match counts.

    Explore this capability — Meeting reading
  6. Settings and version

    Find version, updates and diagnostics in About.

    Recording settings put audio sources first; About brings together version information, update checks and diagnostics.

    Explore this capability — Settings and version

Not on the roadmap

Currently outside the scope

These features are not part of the current roadmap. Adding them would need a separate review of user needs and scope.

  • Habit, health or lifestyle tracking
  • Social sharing of your knowledge
  • AI roles that imitate real, named people
  • Always-on collection of your accounts or activity
  • A cloud copy of your library on our servers

Want to know when it ships?

Drop us a note and we will tell you as soon as there is an installer for your system.

Email us