Model upgrade defaults

[lisafast]

Summary | Résumé

Finished first model upgrade, now need to rationalize all the menus etc.

PROBLEM PROMPT: how to rationalize, across the entire app, how, when we upgrade to a new
  model, do we change the default graph to the new model? In Settings - I've
  now switched to GPT5one - we need a way now and in the future to push the old model to the bottom
  of the list (no longer labelled default) and the new to the default status. That said, this
  worked well during the process of testing the new model, in the live chat and in batches. 3 weeks
  ago I set the default for staging/cdssandbox in settings to GPT5one - then a week ago I flipped
  the settings for prod. And who knows, it's possible we'd have to flip back some day. In the
  meantime though, we need a rational process for the change. Also the workflow in Settings
  overrules the one for live chat from admin with options - notice how in the final screenshot it
  looks as if it is using defaultGraph of 4.1 but when I run a chat, the logs show it's using 5.1.
  So basically, this approach is now broken and needs a refactor. Note this is our first model
  upgrade (from 4.1 to 5.1) so that's why we're identifying all these issues and need a better
  procesS. Come up with a plan with some options - and check how other modern AI apps (haha that
  means 2026 I guess!) do this.

Phase 1 — model.default Setting:

• api/chat/chat-graph-run.js — server resolves model from model.default setting, injects into graph
  input
• src/pages/SettingsPage.js — new "Default model" dropdown
• Locale keys for the label in both languages

Phase 2 — GPT5 graph removal:*

• Deleted GPT5MiniDefaultGraph.js, GPT5OneDefaultGraph.js, GPT5OneChatGraph.js
• agents/graphs/registry.js — legacy aliases map old names → GenericWorkflowGraph
• api/chat/chat-graph-run.js — LEGACY_MODEL_MAP extracts implied model from old graph names
• Removed GPT5* options from all 4 UI dropdowns
• Updated architecture docs

Phase 3 — Single source of truth:

• New config/workflows.js — exports WORKFLOWS, AVAILABLE_MODELS, WORKFLOW_VALUES, MODEL_VALUES
• SettingsPage, ChatOptions, BatchUpload all import and render from the shared config
• Shared workflows.* and models.* locale keys in both languages

After deploy, set workflow.default=DefaultGraph and model.default=openai-gpt51 in Settings. Next
model upgrade: just add to config/workflows.js and flip Settings.

Description en 1 à 3 phrases de la modification proposée, avec un lien vers le
problème (« issue ») GitHub ou la fiche Trello, le cas échéant.

Test instructions | Instructions pour tester la modification

Sequential steps (1., 2., 3., ...) that describe how to test this change. This
will help a developer test things out without too much detective work. Also,
include any environmental setup steps that aren't in the normal README steps
and/or any time-based elements that this requires.

---

Étapes consécutives (1., 2., 3., …) qui décrivent la façon de tester la
modification. Elles aideront les développeurs à faire des tests sans avoir à
jouer au détective. Veuillez aussi inclure toutes les étapes de configuration
de l’environnement qui ne font pas partie des étapes normales dans le fichier
README et tout élément temporel requis.

[github-actions]

🧪 AI Answers Review Environment

https://dmtqwcv4hsag7dccwy3wb6fd7u0hgzkq.lambda-url.ca-central-1.on.aws/