Roadmap

Direction, not dates. Items move between columns as work starts and ships; shipped items link to the changelog and the issue list. Updated 2026-10-07. Fixed issues are on the issue page; release details are in the changelog.

Shipped

Reliability, privacy and permission fixes in the next desktop release. See the issue list for details.

  • WhatsApp reliability
    No crashes on timeouts, sensible reconnect backoff, missed-message notices, health in Settings.
  • Private logs and diagnostics
    Masked identifiers, owner-only files, an application log, and a redacted Export diagnostics bundle.
  • Explore mode that fits real work
    More read-only commands, accurate hints, and Allow once instead of a hard block.
  • Automations you can trust
    Failure alerts, auto-pause, and a Failing jobs view in Mission Control.
  • Safe updates
    Updates wait until the app is idle; interrupted work is marked clearly.
  • Long turns and sleep
    No false session recovery during long thinking; reconnect right after the computer wakes.
  • Provider accuracy
    Correct GitHub Copilot context limits and thinking; Databricks sign-ins refresh and warn before expiry.
  • Public changelog, roadmap and issue list
    This page, the changelog, and a way to report issues.

Next

Work we expect to pick up first.

  • Faster turns
    Run the thinking-level classifier locally, keep the system prompt short and cacheable, and show prompt size per turn.
  • Output styles
    Brief, Standard and Detailed report, remembered per workspace.
  • One approval inbox
    Every pending approval across sessions in one place.
  • Rate-limit awareness
    Show provider limits and wait instead of failing.
  • Global pause and emergency stop
    Stop every running session and automation at once.
  • Accurate context meter
    Show what the model will actually receive, per provider.
  • First-run health check
    Check sign-ins, tools and permissions on first launch.
  • Signed and notarized builds
    Signed installers on macOS and Windows.

Planned

Larger themes we intend to build.

  • Parallel research
    Offer to fan independent questions out to background agents.
  • Steering and undo
    Mid-run steering, branching, and workspace snapshots to undo agent edits.
  • Agents, memory and projects
    Persistent named agents, shared projects, memory you can see and edit, and proactive briefings.
  • Safety and trust
    A credential vault, risk-tiered autonomy, defense against instructions hidden in data, an audit trail and signed extensions.
  • Reach
    More messaging channels, voice in and out, a mobile companion, and notifications with quiet hours.
  • Pages and data
    Living reports, a data-warehouse native agent, a dataset explorer, and workspace export and import.
  • Operations
    In-app documentation search, clearer error messages, resource budgets per session.
  • Transparency
    An explain-what-you-did digest, dry-run mode for automations and skills, and deterministic replay.