Skip to content

Fix/20260407#825

Closed
paullizer wants to merge 413 commits intomainfrom
fix/20260407
Closed

Fix/20260407#825
paullizer wants to merge 413 commits intomainfrom
fix/20260407

Conversation

@paullizer
Copy link
Copy Markdown
Contributor

  • Fact Memory Instructions and Facts
    • Added a clearer Fact Memory experience that distinguishes always-on Instructions from relevance-based Facts on the profile page and in chat-time recall.
    • Chat responses now surface saved-memory usage more clearly through separate Instruction Memory and Fact Memory Recall thoughts and citations.
    • Admin Settings Latest Features and the user-facing Support > Latest Features page now include Fact Memory guidance and screenshots, and admins can show or hide that announcement from General > User-Facing Latest Features.
    • (Ref: semantic_kernel_fact_memory_store.py, route_backend_chats.py, route_frontend_profile.py, profile.html, support_menu_config.py, admin_settings.html, latest_features.html, fact memory guidance and latest-features coverage)

paullizer and others added 30 commits January 24, 2026 06:41
Reverted route_backend_settings.py to origin/development version and removed
AZURE_AI_SEARCH_TEST_CONNECTION_FIX.md documentation. These changes will be
submitted in a dedicated PR to keep the ServiceNow integration PR focused.
…factory.py

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
…RGING_FIX.md

Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Replaced REST API approach with SearchIndexClient SDK to properly handle managed identity authentication in Azure public cloud. The SDK automatically handles token acquisition and endpoint construction, eliminating the 'search_resource_manager is not defined' error that occurred with the REST API approach.
Replaced REST API approach with SearchIndexClient SDK to properly handle managed identity authentication in Azure public cloud. The SDK automatically handles token acquisition and endpoint construction, eliminating the 'search_resource_manager is not defined' error that occurred with the REST API approach.
* updated the logging logic when running retention delete with archiving enabled (#642)

* Corrected version to 0.236.011 (#645)
* updated the logging logic when running retention delete with archiving enabled (#642)

* Corrected version to 0.236.011 (#645)

* v0.237.001 (#649)
* release note updating for github coplilot

* fixed logo bug issue

* added 2,3,4,5,6,14 days to rentention policy

* added retention policy time updates
paullizer and others added 28 commits March 23, 2026 17:24
…260323

Fix/release notes merge fix 20260323
…nute timeout and allow warning dialog to be disabled.
* add limition to py

* rmv mjs node

* add agent templating poc

* upd pipeline

* add agent template gallery and start of ai foundry agents

* push for paul

* add conversationid query param

* move abp docs

* add plugin/tool auth definitions

* upd lang instructions + fix

* adad web search--upd/fixes for control center

* upd pipelines

* add swagger

* upd to pipelines

* add max lengths checks for payload and templates

* fix missing swagger import

* add code doc for behavior

* add submittedAt back

* rmv dup key vault tab

* fix security issue with cross-contamination

* dev publish for sc cc

* upd for sc cc

* add adjustable left nav

* add docs

* add admin configured chunking

* add hide agent templating when not enabled

* add loader changes

* add keepalive in streaming

* fix except + multiendpoints integration + enh thoughts + core plugin thoughts

* add keyvault to multiendpoints. stage new_foundry code. upd req.txt.

* fix for agent buttons in workspaces

* upd agent templating to notifications and approval systems

* 1st pass at enhanced control center + chat fixes

* add msgraph improvement + agent selector fixes

* add message reattach + thoughts subsequent message fix

* fix SK to local working 1.41.0

* regress to 1.40.0

* fix for missing dep in SK for protobuf

* fix for thought streaming and reattach

* unify debug print into log_event

* rmv extraneous file

* upd logging and group mgmt insrtuctions or py

* pr fixes

* adds workspace filtering and UI changes to chat dropdowns

* add css fixes for prompt editor in dark mode

* add agent fallback for models that do not support function tool calling + overhaul of chat UI

* removed the stray user_id deletion

---------

Co-authored-by: Bionic711 <nadoyle@microsoft.com>
Co-authored-by: Paul Lizer <paullizer@microsoft.com>
* initial updates

* reconciled deployment code with latest dev

* updated logic for key usage

* fixed enhanced citation

* improved video indexer support for gov and customer

uses the correct api version and removed openai support for gov

* Update release_notes.md

* resolved PRIVATE_DNS_ZONE_CONFIGS and bicep variable for AOAI
* add chat tutorial

* upd chat tutorial

* Update FULL_WIDTH_CHAT_SUPPORT.md

* Fixed 8 files with unescaped Liquid/Jinja2 template syntax that would cause Jekyll build failures

* Update .gitignore

* Updated to v0.239.001

* updated sidebar

* fixed formatting

* latest release

* Update export-conversation.md

* more features

* added tutorial to workspace and expanded chat

* added noted for tutorials

* latest features and bug/feature submission

* registration submission to receive latest release updates and calls

* added images to new features

* added missing swagger

* added swagger route check

* file no longer used

* removed

* improved docker customization for certs documentation

* Update release_notes.md

* developer documentation for running locally

* improved exception handling for feedback and feedback logging

* Updated check_swagger_routes.py so UTF-8 read failures and SyntaxErrors now produce GitHub Actions ::error annotations instead of crashing

---------

Co-authored-by: Bionic711 <nadoyle@microsoft.com>
User idle warning modal and auto logout (clear session) after configurable timeout value is hit
* do not display popup on multi-endpoint disabled to users in workspaces

* fixed model selection for legacy gpt

* fixed entity too large bug for tabular

* Pinned python packages to an explicit version

* updated python package versions

* improve multi tab joins

* multi-tab analysis improvement and citation improvement for large datasets

What we did in part one was move the large raw tabular/tool payloads off the main assistant message and store them as linked child records instead. The core code is in functions_message_artifacts.py. We added helpers to:

Build a compact citation that stays on the assistant message.
Store the full raw citation payload in separate assistant_artifact records, with assistant_artifact_chunk records if the payload is large enough to need chunking.
Rehydrate those raw payloads later when an export or deep inspection needs the full data again.
In practice, that meant the assistant message no longer carried the full tabular function_result blob. It only kept a summary like tool name, compact arguments, counts, and a few sample rows. The heavy payload was externalized with an artifact_id and raw_payload_externalized=True. That is the “large assistant-side payloads outside primary chat items” change you’re remembering.

The chat write path was updated in route_backend_chats.py to persist those artifact records via persist_agent_citation_artifacts(...). The normal conversation/history paths were updated to hide those artifact records from visible chat history with filter_assistant_artifact_items(...) in route_backend_conversations.py and route_frontend_conversations.py. Export paths were updated to stitch the raw payload back together with build_message_artifact_payload_map(...) and hydrate_agent_citations_from_artifacts(...) in route_backend_conversation_export.py.

Why that helped performance and reliability:

It kept the main assistant message much smaller.
It prevented large tabular citations from pushing a single Cosmos message item over size limits.
It reduced the amount of heavy citation data carried around in normal chat reads/history assembly.
It preserved the raw data for export/debugging instead of throwing it away.
A few extra size cuts were part of that same phase:

We compacted stored citation summaries instead of keeping the full raw result on the main message.
We excluded noisy args like user_id, conversation_id, and source from compact tabular citation arguments in functions_message_artifacts.py.
We also removed the duplicate user_message field from assistant message docs, as called out in ASSISTANT_CITATION_ARTIFACT_STORAGE_FIX.md.

* Update release_notes.md

* Update release_notes.md

* Update release_notes.md

* fix plugin id handling

* fix json parsing
* improved document versioning

* revisions now cite correctly

* close

* staged

* improved search

* improved chat citation and added conversation history citation and though

* multi-file support

* added citation replay

* Update release_notes.md

* updated release timeline

* resolved toast
* updated admin features

* added user features

* updated user latest features

* enable support menu update

* fix centralizes JSON-safe scalar normalization

* support_feedback resolves the feedback fields through explicit data-feedback-field hooks

* fixed voice response visual bug
@paullizer paullizer closed this Apr 8, 2026
@paullizer paullizer deleted the fix/20260407 branch April 8, 2026 00:40
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.

9 participants