Add model catalog management and per-step Auto orchestration routing

[Paul Lizer (paullizer)]

Summary

Administrators need reusable descriptions of model strengths, and orchestration needs task-specific selection rather than one model for every step. This adds a richer catalog without creating another endpoint or credential manager.

• Add shared classic and V2 catalog management with strengths, limitations, capabilities, evidence, search, favorites, priorities, and custom-profile creation, duplication, editing, and archival.
• Associate profiles with existing global, personal, and group AI Connections. Built-in facts remain read-only; connection overrides and independently verified capacity metadata remain authoritative.
• Add explicit Auto selection only to V2 orchestration: eligibility and task suitability precede priority and favorites. Classic chat and regular V2 chat remain manual; orchestration can also stay pinned.
• Preserve model bindings through plan revisions, checkpoint recovery, live events, saved steps, and answer attribution. Reauthorize at execution and reject changed technical metadata rather than silently rerouting.

Profiles alone do not make models callable or publish them. Responses-only models are excluded from this Chat Completions path. Auto does not forecast unknown future document/tool sizes or estimate live pricing and latency; existing token guards remain in force.

Linked issue

N/A. Issue creation was deferred by the requester.

Release Notes & Latest Features

• New Feature
• Bug Fix
• UI Enhancement
• Breaking Change
• Internal only

Is this visible to end users?

• Yes
• No

Is this admin-facing (Admin Settings, governance, deployment, config)?

• Yes
• No

Should this become a Latest Feature card?

• Yes
• No
• Already added

Screenshot needed for the card?

• Yes
• No
• Attached

Release notes and a Latest Feature card are not included. Release-note approval was requested but unavailable, so the file was left unchanged.

Version bump

• application/single_app/config.py VERSION third segment bumped, or not needed because this is docs-only
• deployers/version.txt bumped, or not needed because deployers/ was not changed

Application version: 0.261.125 -> 0.261.126. No deployer changes.

Testing / validation

• Combined Python 3.12 regression suite with pinned application dependencies: 224 tests and 1,397 subtests passed, covering catalog APIs/profiles, endpoint capacity and profile save/reopen, Auto execution, model selection, recovery, revisions, and catalog evidence.
• Real Flask/offline-provider coverage verifies different research and answer models, context restoration, live and persisted attribution, changed-capacity/access rejection, and binding-sensitive checkpoint fingerprints. Real-app editor projections pass under normal and optimized Python.
• Browser checks passed: ui_tests/test_model_catalog_management.py, ui_tests/test_model_endpoint_capacity_editor.py, and standalone ui_tests/test_v2_orchestration_composer.py.
• Standalone executor/schema checks and all three functional_tests/route_tests/ policy scripts passed.
• test_docs_app_surface_coverage.py, test_docs_site_quality.py, and git diff --check passed.
• npm --prefix application/v2_ui run build passed; the existing large-bundle warning remains.

Live Azure inference and tenant deployment were not exercised. Provider calls and storage are offline fixtures, not a claim of production readiness.

Documentation

• Release notes updated, or not needed
• Feature documentation updated, or not needed
• Fix documentation updated, or not needed

Added the Model Catalog admin page, model-routing guide, and versioned feature documentation. Updated AI Models/orchestration/chat-control references and the generated application-surface inventory. No separate fix document is needed for this feature.

Security checklist

• New Flask routes include @swagger_route(security=get_auth_security())
• Settings sent to non-admin frontends use sanitize_settings_for_user()
• Browser JavaScript is served from local SimpleChat static assets only; no CDN-hosted JS
• No secrets, keys, connection strings, or local-only artifacts are included

Catalog-specific API responses use explicit safe projections rather than raw settings. Writes use the existing fenced settings store with ETag conflict handling; model selection stays within authorized connections.