Skip to content

feat(desktop): bring cloud/local context policy to rel/xp-build - #626

Merged
Ooooze merged 5 commits into
rel/xp-buildfrom
codex/desktop-cloud-context
Oct 8, 2026
Merged

Ooooze merged 5 commits into
rel/xp-buildfrom
codex/desktop-cloud-context

Conversation

@Ooooze

@Ooooze Ooooze commented Oct 8, 2026 •

Copy link
Copy Markdown
Contributor

The desktop branch used bounded local context for every model. Merge main through #625 so cloud policy retains full tool results in a stable conversation journal and compacts near the model window, while local policy keeps the bounded prompt.

The desktop saves provider config independently of the agent's TUI helper. Initialize modelMode for new desktop connections using the same known-service classification, and apply the v75 cloud migration to newly imported providers from older terminal configs. Defaults preserve existing desktop choices, exact-model overrides and current-format inherit. A contract test compares the separate desktop classifier with the actual core implementation and all current provider presets.

This uses a merge commit to retain main ancestry. It does not add a desktop policy picker or publish an app release.

Validation on Node 25.7:

  • Full core test:ci: 1,123 files, 13,629 passed, 4 skipped.
  • Root build, test typecheck (no new diagnostics), import boundaries and documentation checks passed.
  • Full desktop build, including the macOS speech helper, passed.
  • All 170 desktop unit tests passed, including new connection, import, policy preservation and core/desktop parity checks.
  • Windows desktop smoke passed on x64 and ARM64 with the repository's existing known-failure exclusions.

Live Electron chat with real local/cloud models was not exercised.

sosidudku1 and others added 5 commits October 7, 2026 17:33
* feat(local-models): keep selection responsive during engine startup

* fix(local-models): allow cold GPU setup on interactive starts

* test(local-models): cover readiness and cancellation after cold startup

* feat(tui): accept messages while a local model starts

Show the accepted prompt and a delay warning immediately. Wait for the final model choice with cancellation and original-session ownership before inference, and keep failed or stopped starts from dispatching a waiting turn.

* fix(tui): retain queued messages after model startup fails

Do not drain accepted follow-ups into a failed configuration. Check readiness on queued dispatch and resume the retained backlog after a successful model start.

* fix(tui): scope waiting messages to their session and route

Keep failed local startup state from affecting cloud follow-ups. Leaving a chat cancels its retained idle backlog with the existing queue notice instead of allowing those messages to follow the selected session.

---------

Co-authored-by: Valera Brizhatiuk <19537764+plombeer31@users.noreply.github.com>
@Ooooze
Ooooze merged commit 167a87d into rel/xp-build Oct 8, 2026
2 checks passed
sosidudku1 added a commit that referenced this pull request Oct 12, 2026
…rkspace skills) into the Atomic Core desktop branch
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants