fix(inference): include model field in nvidia-nim API requests (#2784)

[sanil-23]

Closes #2784

Root cause

In make_cloud_provider_by_slug (factory.rs), when a provider string like "nvidia-nim:" was used (slug with empty model after the colon), the effective model resolution fell through to entry.default_model.clone().unwrap_or_default(). If the cloud provider entry had no default_model configured, this produced an empty string "", which was then sent as the model field in the OpenAI-compatible /v1/chat/completions request body, causing a 400 error ("model field is required").

Fix

Added a two-level fallback chain in make_cloud_provider_by_slug:

1. Try config.default_model (the user's top-level default model setting)
2. Fall back to the global DEFAULT_MODEL constant (reasoning-quick-v1)

This mirrors the same fallback pattern already used by cloud_entry_provider_string (the sibling function that builds provider strings for the primary cloud route).

Verified

• cargo check -p openhuman — compiles cleanly
• All 78 factory unit tests pass
• All 91 compatible-provider unit tests pass

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 8fda9b7d-48d6-4132-96de-f1211cf2989d

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

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

[oxoxDev]

Author: @sanil-23 (MEMBER — core team)

⚠️ Duplicate of in-flight PR #2791 by @graycyrus — same root cause, same function (make_cloud_provider_by_slug in src/openhuman/inference/provider/factory.rs), identical title. #2791 was opened 11 days earlier and already has coderabbit approval. Issue #2784 is assigned to @graycyrus.

Surfacing as COMMENT (not blocking with CHANGES_REQUESTED) so you can coordinate with @graycyrus on which approach to land:

Comparison

Aspect	#2791 (graycyrus)	#2824 (this PR)
Files	factory.rs + config_rejection.rs + factory_test.rs	factory.rs only
Fallback chain	entry.default_model → bail with actionable error	entry.default_model → config.default_model → DEFAULT_MODEL constant (silent)
Sentry guard	Adds "model field is required" to config_rejection classifier (suppresses TAURI-RUST-4NM)	None
Tests	3 unit tests	None
AuthStyle::OpenhumanJwt exemption	Yes	No
Failure mode	Loud, surfaces config bug	Silent — masks misconfig

Concerns (independent of duplicate question)

1. factory.rs:779 — silent DEFAULT_MODEL fallback masks misconfiguration. reasoning-quick-v1 is openhuman's abstract tier (Kimi K2.6 Turbo on Fireworks), NOT an nvidia-nim-native model. When the chain hits the third arm with slug=nvidia-nim, the request hits nvidia-nim's /v1/chat/completions with model=reasoning-quick-v1 — model nvidia-nim does not host → 404. Replaces a 400 with a different cryptic 4xx; user has zero signal their config is wrong.

2. No regression tests. Coverage gate (≥80% on changed lines) and #2784's AC ("Unit test added to verify model field is present in nvidia-nim request payloads") — neither met.

3. Missing AuthStyle::OpenhumanJwt guard. make_openhuman_backend derives its model from config.default_model independently — the third-arm DEFAULT_MODEL rescue is dead code for openhuman entries and masks an honest empty for misconfigured auth_style cases. #2791 gates this with if entry.auth_style != AuthStyle::OpenhumanJwt && effective_model.trim().is_empty().

4. No config_rejection.rs update. TAURI-RUST-4NM Sentry events from older clients with bad configs will keep firing post-deploy until install base rotates. #2791 adds "model field is required" to is_provider_config_rejection_message precisely to silence this.

Recommendation

Close #2824 in favour of #2791 (more complete, has tests, fails loudly, by ticket assignee). If you want the global-DEFAULT layer for some other reason, add it onto #2791's tail. Per [[feedback_check_open_prs_before_implement]] — checking open PRs against #2784 would have caught the overlap before opening.

Happy to flip to APPROVE / CHANGES_REQUESTED once you decide. Not blocking with formal review state yet.

Verified / looks good

• .filter(|m| !m.trim().is_empty()) guards correctly handle whitespace-only default_model strings — actual hardening over main's unwrap_or_default() which only handles None.
• crate::openhuman::config::DEFAULT_MODEL path resolves (src/openhuman/config/schema/types.rs:37).
• CI green (2 e2e + 1 coverage-gate pending), no formatting/clippy issues.

[sanil-23]

Thanks for the side-by-side @oxoxDev — agreed, #2791 is the right landing (loud bail!, tests, AuthStyle::OpenhumanJwt guard, config_rejection.rs silencer for TAURI-RUST-4NM). The silent DEFAULT_MODEL fallback here would just trade a 400 for a 404 against nvidia-nim's /v1/chat/completions.

Closing in favor of @graycyrus's #2791. Should have checked open PRs against issue #2784 before opening — apologies for the noise.