slurpojoe bolts on a cheaper AI engine and a way around OpenAI's meter

This fork adds a fourth AI model option and a trick to run OpenAI calls off an existing ChatGPT login instead of a paid API key.

infrastructureintegration

slurpojoe is chasing two things at once: lower inference costs and a way to stop paying OpenAI per call. The first move is clean and low-risk - it adds DeepSeek, a lower-cost AI model maker, as a fourth engine option alongside the existing providers, complete with streaming, tool use, and a tested wiring job. If you just want a cheaper model on the menu, this part is easy to lift.

The second move is the spicy one. Instead of a paid OpenAI API key, the fork quietly reuses a local ChatGPT/Codex desktop login to authorize calls, refreshing the token in the background. Clever, but worth a hard look before anyone copies it: it leans on a setup that may run against OpenAI's terms of service, could break if OpenAI changes anything, and only works for a single self-hosted instance, not a multi-user product.

So what Anyone weighing the running cost of a legal-AI deployment should look - one half is a safe cost cut, the other is a gamble dressed as one.

View this fork on GitHub →

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

Commits in this thread

4 commits from slurpojoe/make, oldest first. Source extracted verbatim from the harvested git log.

SHA	Subject	Author	Date
`a3370b11`	Add design spec: DeepSeek provider + Codex OAuth integration	slurpojoe	2026-05-28	↗ GitHub
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
`b0b317aa`	Revise design spec with verified Codex OAuth and additional fixes	slurpojoe	2026-05-28	↗ GitHub
commit body - Verified Codex refresh endpoint and JSON body format against openai/codex - DeepSeek env-only (avoids DB migration for user_api_keys CHECK constraint) - Add type extensions (UserApiKeys, ApiKeyState, Provider unions) - Tilde expansion, atomic file write, in-flight refresh locking - Thinking parameter only for deepseek-v4-pro - Switch-based router with exhaustive check - Default model fallback when stored/default is unavailable Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
`8b2e0517`	feat: add DeepSeek provider and Codex OAuth token support	slurpojoe	2026-06-05	↗ GitHub
commit body - New: deepseek.ts - Chat Completions streaming, tool calls, thinking for v4-pro - New: codexAuth.ts - JWT decode, auto-refresh, concurrent-refresh dedup, atomic write - Extended: types, models, exhaustive switch dispatch, userApiKeys, user routes - Frontend: DeepSeek group in ModelToggle, modelAvailability, ApiKeyState, account settings - UX: useSelectedModel falls back to first available model if default unconfigured - Secrets: allowedDevOrigins moved from next.config.ts to NEXT_DEV_ALLOWED_ORIGINS env var - Tests: vitest setup + 9 tests covering codexAuth and deepseek error paths - Fix: ChatInput.tsx declaration order (apiKeys used before declaration) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
`6d67aa19`	fix: flush pendingText on maxIter exhaustion; remove dead ModelToggle exports	slurpojoe	2026-06-05	↗ GitHub
commit body - deepseek.ts: hoist pendingText before loop + flush after loop ends to prevent silent data loss when tool-call iteration limit is reached - deepseek.ts: annotate "developer" role as unverified for v4-flash - ModelToggle.tsx: remove DEEPSEEK_MID/LOW_MODELS (unused dead code; backend/src/lib/llm/models.ts is the authoritative source) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

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-642.md from inside the repo you want the changes in.

⬇ Download capture-thread-642.md