Cpodlaskilegal adds OpenAI, then hands it the keys

This fork now runs three AI engines, and the new one just became the default.

infrastructureintegration

Mike ships with two AI providers built in: Anthropic's Claude and Google's Gemini. Cpodlaskilegal has added a third, OpenAI, the company behind ChatGPT and the GPT models. It is a full integration, not a stub: streamed responses, tool use, and a dial that spends more reasoning effort on the heavyweight models and less on the small, cheap ones. Users can pick OpenAI models from the same menus where Claude and Gemini already live.

Then came the second move. A follow-up change makes OpenAI the default engine across the app, with the largest GPT model handling main conversations and smaller ones covering quick jobs like naming chats and reading tables. Gemini was pulled from the model menus entirely. The setup leans on an OpenAI key managed on the server, so the firm supplies the key rather than each user.

So what If your firm's model preferences (or contracts) point to OpenAI rather than Google or Anthropic, this fork shows what that switch looks like in practice.

View this fork on GitHub →

Spotted something wrong? Or know the PR text has fresher detail than the writeup above?

Commits in this thread

2 commits from Cpodlaskilegal/mike, oldest first. Source extracted verbatim from the harvested git log.

SHA Subject Author Date

7fbbe5c9 Add OpenAI model provider Cpodlaskilegal 2026-05-01 ↗ GitHub

2d0a0e10 Fix Azure auth + retire Gemini defaults Cpodlaskilegal 2026-05-06 ↗ GitHub

SHA	Subject	Author	Date
`7fbbe5c9`	Add OpenAI model provider	Cpodlaskilegal	2026-05-01	↗ GitHub
`2d0a0e10`	Fix Azure auth + retire Gemini defaults	Cpodlaskilegal	2026-05-06	↗ GitHub
commit body Auth - Backend pinned the v2.0 issuer but the Entra API app was still on v1 tokens (accessTokenAcceptedVersion=null), so every real MSAL token failed jwtVerify with 'Invalid or expired token'. The API app has now been flipped to v2 in Azure (Graph PATCH api.requestedAccessTokenVersion). Models - Default main/title/tabular models switched from Gemini to OpenAI equivalents that work with the Azure-managed OPENAI_API_KEY: main -> gpt-5.5 title -> gpt-5.4-nano tabular -> gpt-5.4-mini - user_profiles.tabular_model column default updated in both schema files (azure_postgres_schema.sql, 000_one_shot_schema.sql) and ALTERed in the live database. - Frontend fallbacks in ModelToggle, UserProfileContext, account/models page, TRChatPanel, and TabularReviewView updated accordingly. - Gemini stripped from the assistant + tabular model dropdowns and from the API Keys section of the account/models page; ModelOption.group type tightened to 'OpenAI' \| 'Anthropic'. - resolveTitleModel now prefers the server-managed OpenAI key, then a user Claude key, then a user Gemini key (was Gemini-first).

commit body

Auth
- Backend pinned the v2.0 issuer but the Entra API app was still on v1
  tokens (accessTokenAcceptedVersion=null), so every real MSAL token
  failed jwtVerify with 'Invalid or expired token'. The API app has now
  been flipped to v2 in Azure (Graph PATCH api.requestedAccessTokenVersion).

Models
- Default main/title/tabular models switched from Gemini to OpenAI
  equivalents that work with the Azure-managed OPENAI_API_KEY:
    main    -> gpt-5.5
    title   -> gpt-5.4-nano
    tabular -> gpt-5.4-mini
- user_profiles.tabular_model column default updated in both schema files
  (azure_postgres_schema.sql, 000_one_shot_schema.sql) and ALTERed in the
  live database.
- Frontend fallbacks in ModelToggle, UserProfileContext, account/models
  page, TRChatPanel, and TabularReviewView updated accordingly.
- Gemini stripped from the assistant + tabular model dropdowns and from
  the API Keys section of the account/models page; ModelOption.group
  type tightened to 'OpenAI' | 'Anthropic'.
- resolveTitleModel now prefers the server-managed OpenAI key, then a
  user Claude key, then a user Gemini key (was Gemini-first).

Capture this thread into my fork

Download a single Markdown prompt that tells Claude how to port every commit above into your working tree — adapting paths and structure to match your repo. Run it via claude -p < capture-thread-819.md from inside the repo you want the changes in.

⬇ Download capture-thread-819.md