fix(inference): propagate anthropic wire mode to Hermes sandbox config sync

[jahubba]

Summary

nemoclaw inference set --provider anthropic-prod (or compatible-anthropic-endpoint) leaves Hermes-type sandboxes unable to run any inference: the Hermes config sync writes model.provider: custom without an api_mode, so the agent speaks OpenAI chat-completions at an anthropic-type provider and the gateway L7 policy denies every call (DENIED inference.local:443 POST /chat/completions). This PR propagates the anthropic wire mode through the Hermes sync, matching what patchOpenClawInferenceConfig already does for OpenClaw sandboxes.

Related Issue

Fixes #4746

Changes

• src/lib/actions/inference-set.ts — patchHermesInferenceConfig writes model.api_mode: anthropic_messages when route.inferenceApi === "anthropic-messages" (Hermes honors this key for custom providers), and clears the key on non-anthropic routes so a stale wire mode never survives a provider switch.
• src/lib/inference/config.ts — getSandboxInferenceConfig keeps compatible-anthropic-endpoint on the OpenAI wire only for an explicit preferredInferenceApi === "openai-completions"; the null-default no longer silently selects the OpenAI branch for an anthropic-type provider (the Hermes sync never passes a preference).
• 6 new unit tests across inference-set.test.ts and config.test.ts: anthropic wire propagation for both providers, stale api_mode cleanup on switch-away, anthropic-prod route mapping, default-vs-explicit preference for compatible-anthropic-endpoint.

Note on local verification: this was developed on an aarch64 Jetson host where a fixed set of environment-dependent suites (NIM/GPU preflight/install checks) fail identically on a clean main checkout; the affected test files pass 51/51 and npm run typecheck:cli + Biome are clean. Relying on CI for the full matrix.

Type of Change

• Code change (feature, bug fix, or refactor)
• Code change with doc updates
• Doc only (prose changes, no code sample modifications)
• Doc only (includes code sample changes)

Verification

• npx prek run --all-files passes
• npm test passes
• Tests added or updated for new or changed behavior
• No secrets, API keys, or credentials committed
• Docs updated for user-facing behavior changes
• npm run docs builds without warnings (doc changes only)
• Doc pages follow the style guide (doc changes only)
• New doc pages include SPDX header and frontmatter (new pages only)

---

Signed-off-by: Jason Hubbard jasonahubbard@gmail.com

Summary by CodeRabbit

Release Notes

• Bug Fixes

  • Fixed Anthropic endpoint routing to correctly respect explicit API preference selections and prevent unintended protocol switches when using compatible endpoints.

• Tests

  • Expanded test coverage for Anthropic endpoint configuration and provider switching scenarios.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: afe2f4c0-2160-44ac-a026-8a372b81fbb0

📥 Commits

Reviewing files that changed from the base of the PR and between 17734b1 and 37760bc.

📒 Files selected for processing (4)

• src/lib/actions/inference-set.test.ts
• src/lib/actions/inference-set.ts
• src/lib/inference/config.test.ts
• src/lib/inference/config.ts

---

📝 Walkthrough

Walkthrough

This PR fixes Hermes sandbox inference for Anthropic providers by adding the missing anthropic-messages wire-mode declaration to the Hermes config and tightening the conditional for compatible-anthropic-endpoint to avoid silent OpenAI-wire fallback when no API preference is passed.

Changes

Hermes Anthropic Provider Wire-Mode Fix

Layer / File(s)	Summary
Hermes config api_mode propagation src/lib/actions/inference-set.ts, src/lib/actions/inference-set.test.ts	patchHermesInferenceConfig now sets modelConfig.api_mode: "anthropic_messages" when the resolved inference API is anthropic-messages, and removes stale api_mode for all other APIs. Tests verify propagation for both anthropic-prod and compatible-anthropic-endpoint, and verify cleanup when switching from Anthropic to OpenAI routes.
Sandbox inference config conditional tightening src/lib/inference/config.ts, src/lib/inference/config.test.ts	getSandboxInferenceConfig now requires explicit preferredInferenceApi === "openai-completions" to route compatible-anthropic-endpoint to the OpenAI wire, preventing silent fallback when no preference is passed. Tests verify anthropic-prod routes to Anthropic wire, compatible-anthropic-endpoint defaults to Anthropic wire, and switches to OpenAI wire only on explicit openai-completions request.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Suggested labels

fix, enhancement: inference, area: inference, integration: hermes, bug-fix, area: providers

Suggested reviewers

• cv

Poem

🐰 A Hermes hop through Anthropic fields,
Where api_mode now reveals its yields,
No silent falls to OpenAI's way,
Just anthropic-messages on display,
Configuration aligned, the agents play! 🎯

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and specifically describes the main change: propagating anthropic wire mode to Hermes sandbox config sync, which is the core fix for the broken Hermes-Anthropic inference.
Linked Issues check	✅ Passed	The PR fully implements both required coding fixes from issue #4746: (1) patchHermesInferenceConfig now writes model.api_mode for anthropic-messages and clears it for non-anthropic routes; (2) getSandboxInferenceConfig restricts compatible-anthropic-endpoint to OpenAI wire only on explicit openai-completions preference.
Out of Scope Changes check	✅ Passed	All changes are directly scoped to fixing the two root causes identified in issue #4746 across inference-set.ts, config.ts, and their corresponding test files with no extraneous modifications.

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

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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

[sandl99]

#4402 did fix the issue on v0.0.59

[jahubba]

Thanks @sandl99. I think #4847 is needed to fully resolve this one. The path this issue reports — nemoclaw inference set --provider anthropic-prod — is handled by patchHermesInferenceConfig, which on v0.0.59 still writes provider: custom with no api_mode, so the gateway keeps denying /chat/completions. #4847 adds the missing api_mode to that patch and lands in v0.0.60 (verified: its patchHermesInferenceConfig → Anthropic-routes test asserts api_mode: anthropic_messages). So this is resolved as of v0.0.60. 👍