Skip to content

Staging#231

Merged
ahmad-ajmal merged 95 commits intomainfrom
staging
Apr 28, 2026
Merged

Staging#231
ahmad-ajmal merged 95 commits intomainfrom
staging

Conversation

@ahmad-ajmal
Copy link
Copy Markdown
Collaborator

No description provided.

AlanAAG and others added 30 commits April 14, 2026 20:04
…model

- Merge generate_multimodal_multi_image into generate_multimodal (image_bytes_list param)
- Add json_mode param to describe_image_bytes; describe_image_ocr now a thin wrapper
- understand_video pulls model from get_vlm_model() with gemini-1.5-pro fallback
- Add test suites: gemini_client_multimodal, vlm_json_mode, ocr_wrapper, understand_video_model
…tion now workcorrectly end-to-end. Issue fixed.

Fixes:
1. STT Language switching screens issue fixed
2. Model fails issue fixes with auto retry, and also after auto retry, chat messages appear: the error text + a "What would you like to do?" message with Retry / Change Model buttons.
3. Added provider-switching directions work correctly end-to-end (Between provider LLLMs& Local LLMs ).  Made full chain: Frontend save, backend validate, settings write,  cache invalidate.
Compress old events, keep recent ones verbatim, the agent always has fresh, recent context with no file rename and no prompt changed no cognitive load increase.

Key point is: old events near the top of the file are already low-value history. The agent cares most about what just happened, and those lines are always preserved like a rolling memory.

- No .old.md file is ever created  so  no prompt changes needed...

- EVENT.md stays as EVENT.md so agent reads it exactly as before

- When file hits 10 MB, we drop oldest 1/3, keep newest 2/3 in place so Agent always sees the most recent events so  no cutoff !
…xt area

- User can attach files
- Drag and drop support
- Copy and paste support
- User can click on the attached file to preview it
- User can also remove the file when not needed
memory issues fixed
zfoong and others added 29 commits April 24, 2026 16:46
- UIController: run react() directly on main loop instead of a worker-thread
  loop; log start/exit and re-raise CancelledError instead of swallowing.
- action/manager: add asyncio.wait_for compat shim for nest_asyncio 1.6.0
  on 3.11+ (band-aid; remove once nest_asyncio is refactored out).
Replace the LLM-only routing in _handle_chat_message with layered
deterministic rules (UI reply target, reply marker, single waiting task)
and fall back to a conservative routing LLM only for ambiguous cases.
Drops the SINGLE-ACTIVE-SESSION-BIAS rule that caused notifications
and unrelated chats to leak into active tasks.

Also bundles two related fixes:
- Preserve recent events during summarization so the UI sees action_end
  before it's purged (action items no longer stuck on "running")
- Propagate status=error when an atomic action throws (run_python no
  longer reports "completed with status: success" after exceptions)
Split the release into a small installer EXE that fetches an agent zip
from GitHub Releases at install time. Wizard rewritten with pywebview + HTML/CSS/JS for native Win11 / macOS styling. Move spec files + installer deps into packaging/; data dir moves to per-user LOCALAPPDATA so installs survive across runs.
@ahmad-ajmal ahmad-ajmal merged commit e2fd902 into main Apr 28, 2026
0 of 2 checks passed
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.

4 participants