fix(inference): point MiniMax at its OpenAI-compatible /v1 endpoint (#3401)

[oxoxDev]

Summary

openhuman.inference_list_models fired provider returned 404: 404 page not found for the MiniMax builtin provider (Sentry TAURI-RUST-8X3: 236 events / 30 users, live on 0.57.13, all platforms). Root cause: MiniMax's catalog entry pointed at its Anthropic Messages base (…/anthropic) with Anthropic auth, but OpenHuman only speaks the OpenAI chat-completions dialect — so the probe built …/anthropic/models, a path MiniMax doesn't serve. Fix: point MiniMax at its OpenAI-compatible /v1 surface with Bearer auth.

Problem

list_configured_models (src/openhuman/inference/provider/ops.rs) builds {endpoint}/models, and the chat provider (compatible.rs) builds {endpoint}/chat/completions — both OpenAI-style paths. OpenHuman has no Anthropic Messages provider; AuthStyle::Anthropic only swaps the auth header, it still hits /chat/completions and /models.

MiniMax was added in #3062 with endpoint = https://api.minimax.io/anthropic + AuthStyle::Anthropic. That base only serves the Anthropic Messages protocol (/anthropic/v1/messages), so with OpenHuman's OpenAI paths both endpoints 404:

URL OpenHuman builds	Result
…/anthropic/models (list_models)	404 page not found ← the Sentry bug
…/anthropic/chat/completions (chat)	404 page not found (chat was broken too)

MiniMax's actual OpenAI-compatible surface (verified by live probe) does serve both:

URL	Result (unauthenticated)
https://api.minimax.io/v1/models	401 — exists, Bearer auth
https://api.minimax.io/v1/chat/completions	401 — exists, Bearer auth

This 404 is exactly the class of failure #2959 ("revert: remove all Sentry error suppression") intended to surface for a root-cause fix, not re-suppress — so the fix corrects the misconfiguration rather than swallowing the error.

Solution

Correct the MiniMax catalog entry (Rust BUILTIN_CLOUD_PROVIDERS + the frontend mirror builtinCloudProviders.ts, kept in sync):

• endpoint: https://api.minimax.io/anthropic → https://api.minimax.io/v1
• auth_style: Anthropic → Bearer

Now list_models → …/v1/models ✓ and chat → …/v1/chat/completions ✓. No change to the probe's 404 handling: a genuine 404 from any provider still errors and reaches Sentry, per #2959.

No suppression, no probe-skipping, no synthetic empty result — the failing operation now actually succeeds.

Submission Checklist

• Tests added or updated (happy path + at least one failure / edge case) per Testing Strategy
• Diff coverage ≥ 80% — changed catalog lines are asserted by the Rust migration/preset tests and the frontend builtinCloudProviders.test.ts + AIPanel.test.tsx MiniMax-connect test
• Coverage matrix updated — N/A: configuration correction, no feature added/removed/renamed
• All affected feature IDs from the matrix are listed under ## Related — N/A (config correction)
• No new external network dependencies introduced
• N/A: Manual smoke checklist — release-cut surfaces untouched (catalog data correction only)
• Linked issue closed via Closes #NNN in the ## Related section

Impact

• MiniMax model-listing now resolves (/v1/models) — AI-settings dropdown populates and save succeeds.
• MiniMax chat, also 404 before, now works (/v1/chat/completions).
• Sentry TAURI-RUST-8X3 stops firing — the endpoint is correct, not silenced. Genuine listing failures from other providers still error → Sentry → root-cause fix.
• Zero regression: MiniMax was fully non-functional before (both paths 404), so this can only fix it.
• Scope: data correction only — no code-path, schema, or error-handling change.

Note: a quick sweep of the other catalog providers found their /models endpoints resolve (401/200); MiniMax was the structural outlier. (google's OpenAI-compat /models 404s unauthenticated but is key-gated — unconfirmed in this Sentry issue; left for separate verification.)

Related

Closes #3401
Sentry-Issue: TAURI-RUST-8X3

AI Authored PR Metadata (required for Codex/Linear PRs)

Linear Issue

N/A — GitHub issue #3401 (from Sentry TAURI-RUST-8X3).

Commit & Branch

Branch fix/3401-list-models-discovery; single commit correcting the MiniMax catalog entry (Rust + frontend mirror) and updating the affected tests.

Validation Run

• pnpm --filter openhuman-app format:check (prettier — changed files clean)
• pnpm typecheck
• Focused tests: cargo test --lib openhuman::config::schema::cloud_providers (4 pass); Vitest builtinCloudProviders.test.ts + AIPanel.test.tsx (45 pass)
• Rust fmt/check (changed): cargo fmt --check, cargo check, cargo clippy clean
• N/A: Tauri fmt/check — shell unchanged

Validation Blocked

None — pushed with the pre-push hook enabled (vendored submodules initialized in the worktree).

Behavior Changes

• MiniMax provider now uses https://api.minimax.io/v1 + Bearer; model-listing and chat resolve. No change to error handling for any other provider.

Parity Contract

Rust catalog and the frontend builtinCloudProviders.ts mirror are updated together and asserted by tests on both sides.

Duplicate / Superseded PR Handling

Supersedes this PR's own earlier approach (a 404-reclassification), which was withdrawn because PR #2959 explicitly reverted that style of Sentry-error suppression. This revision fixes the root-cause misconfiguration instead.

Summary by CodeRabbit

• Chores
  • MiniMax cloud provider configuration has been updated to use OpenAI-compatible API endpoints with bearer token authentication (https://api.minimax.io/v1). Replaces previous Anthropic-style endpoint. Improves API standardization and aligns provider integration with modern authentication patterns, ensuring better consistency across the platform's cloud provider ecosystem.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 8601d899-1b8d-4d9a-951b-de2369c793ac

📥 Commits

Reviewing files that changed from the base of the PR and between 298c3e5 and 24bf14b.

📒 Files selected for processing (5)

• app/src/components/settings/panels/__tests__/AIPanel.test.tsx
• app/src/components/settings/panels/__tests__/builtinCloudProviders.test.ts
• app/src/components/settings/panels/builtinCloudProviders.ts
• src/openhuman/config/schema/cloud_providers.rs
• tests/config_auth_app_state_connectivity_e2e.rs

---

📝 Walkthrough

Walkthrough

MiniMax cloud provider configuration is switched from Anthropic-style endpoint and authentication to OpenAI-compatible /v1 endpoint with Bearer auth across frontend and backend systems, with test expectations updated throughout.

Changes

MiniMax Provider OpenAI-Compatible API Migration

Layer / File(s)	Summary
Frontend MiniMax provider configuration and unit tests app/src/components/settings/panels/builtinCloudProviders.ts, app/src/components/settings/panels/__tests__/builtinCloudProviders.test.ts	MiniMax entry in built-in cloud providers now specifies /v1 endpoint (https://api.minimax.io/v1) with bearer auth style, replacing the prior /anthropic endpoint and anthropic auth style, with added comments explaining the prior 404 behavior and sync requirements with Rust catalog.
Frontend AIPanel integration test app/src/components/settings/panels/__tests__/AIPanel.test.tsx	MiniMax provider connection test renamed and assertions updated to verify the newly configured OpenAI-compatible /v1 endpoint and Bearer auth style in the provider connection flow.
Backend MiniMax provider configuration and defaults test src/openhuman/config/schema/cloud_providers.rs	MiniMax built-in provider entry updated to use OpenAI-compatible /v1 endpoint with AuthStyle::Bearer (replacing /anthropic and AuthStyle::Anthropic), and phase-one defaults test case updated to match the new endpoint and auth style expectations.
Backend E2E connectivity test tests/config_auth_app_state_connectivity_e2e.rs	Minimax legacy migration test assertions updated to expect the OpenAI-compatible /v1 endpoint and Bearer auth style instead of the prior Anthropic-style configuration.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related issues

• Inference: list_models 404s for non-OpenAI builtin providers (minimax/Anthropic) — blocks save + Sentry flood #3401: Root cause fix for MiniMax API list_models endpoint 404 errors; these changes update the provider endpoint from the incorrect /anthropic/models path to the correct OpenAI-compatible /v1/models endpoint with proper Bearer authentication.

Poem

🐰 A whisker-twitch of joy!
From Anthropic ways to OpenAI's bright door,
MiniMax now speaks the /v1 tongue,
Bearer auth flows like morning clover dew—
Test stacks align, the rabbits sing! 🌟

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately and specifically summarizes the main change: updating MiniMax to use its OpenAI-compatible /v1 endpoint instead of the anthropic endpoint, which is the core fix across all modified files.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[sanil-23]

Approve ✅

Traced how endpoint + auth_style actually flow through the inference layer to confirm the root-cause claim holds. The fix is correct, minimal, and well-tested.

Correctness — verified

The comments' root-cause explanation is accurate, and the code confirms it:

• In factory.rs, AuthStyle::Anthropic does not build an Anthropic Messages-protocol client — it builds an OpenAI-compatible provider that still POSTs to {base}/chat/completions and lists {base}/models, only swapping the auth headers to x-api-key + anthropic-version.
• So the old config hit https://api.minimax.io/anthropic/chat/completions and /anthropic/models — neither exists on MiniMax's Anthropic surface (which lives at /anthropic/v1/messages), hence the 404s (incl. TAURI-RUST-8X3).
• /v1 + Bearer now resolves to https://api.minimax.io/v1/chat/completions and /v1/models with Authorization: Bearer — MiniMax's documented OpenAI-compatible surface. Both chat and model-listing now resolve.

Completeness

• Both catalogs (TS builtinCloudProviders.ts + Rust cloud_providers.rs) kept in sync, with comments explicitly calling out the sync requirement.
• Tests updated in all four relevant places (TS unit, TS AIPanel integration, Rust defaults, Rust legacy-migration e2e).
• No collateral: is_minimax_alias is a dead stub; system-role-rejection and <think>-stripping key on model-id patterns, not the endpoint.

One non-blocking note

Already-connected MiniMax users are not auto-migrated. migrate_legacy_fields only fills endpoint when empty and only overrides auth_style when it's still the default Bearer. A user who previously connected MiniMax through the UI has a persisted entry with endpoint=".../anthropic" and auth_style=Anthropic, so neither gets rewritten — those connections stay 404'd until the provider is re-added. New connections and endpoint-less legacy type=minimax configs (covered by the e2e test) do get the fix. Closing the gap would need a one-time migration to rewrite stale built-in endpoints — reasonably out of scope here, but worth a mention in the description or a follow-up.