Skip to content

Great Me

Choose a tag to compare

View all tags
@knoop7 knoop7 released this 07 May 06:33
v7.2.0
c674e10

Fix: Reply Prefix & Language Detection

  • Fix duplicate prefix bug — Switching between Chinese and English no longer produces (deepseekv4) Reply: (deepseekv4) Reply: double prefixes
  • Add reply_formatter.py — Single source of truth for prefix add/strip/language detection, replacing scattered logic across 6 files
  • Fix multi-model comparison mode — All AI replies now correctly labeled with agent name and separated by ---

Conversation Mode Rename

Before After Description
No name Simple Final reply only, no AI attribution
Show AI name Show source Prefixes each reply with the responding AI's name
Full display Multi-model All configured AIs answer simultaneously, each labeled and separated

Anchor-Based Patch Engine (Major Addition)

New text_patch.py — a text patching engine based on anchor matching. AI no longer needs to rewrite entire configs, just locate an anchor and describe the change:

  • 7 operations — replace / insert_before / insert_after / delete / prepend / append / create
  • Precise targeting — Supports literal and regex matching with unique / first / last / all / index selection
  • Atomic guarantees — All patches run sequentially on an in-memory copy; any failure rolls back everything
  • dry_run preview — See the diff before applying, preventing accidental damage
  • Integrated tools:
    • Automation — action=patch for surgical edits to triggers/conditions/actions
    • DashboardCard — Patch card configs without full rewrite
    • Script — Patch individual script steps
    • SearchTools — Patch search configurations

Example: changing an automation trigger time from 08:00 to 09:00 used to require the AI to output the entire YAML. Now it just needs:

patches:
  - op: replace
    anchor: "at: '08:00'"
    new_text: "at: '09:00'"

Thinking Content Support

  • Filter thinking_content from buggy upstream LLM integrations to prevent frontend rendering issues
  • Sanitize pipeline event payloads to block internal data from leaking to the UI

Tool Enhancements

  • Web Search — Improved multi-source result extraction and aggregation
  • AI Collaboration — Cross-agent consultation and parallel tool execution

Runtime Improvements

  • Add resolve_agent_id
  • Correctly resolve third-party agent config_entry_id to entity_id
  • Improve config flow validation and agent selection UX
  • Harden Skill Store / Memory Store async operations
  • Optimize workspace document loading

#7 @rcarmo

Full Changelog: v7.1.0...v7.2.0

Contributors

rcarmo
Assets 2
Loading