feat(providers): add Yandex Cloud AI Studio model provider

[pamnard]

Summary

Add first-class support for Yandex Cloud AI Studio (Model Gallery LLM API). This is the inference backend (llm.api.cloud.yandex.net), not SpeechKit TTS/STT (#27815, #23995).

• Provider id: yandex (aliases: yandex-ai-studio, yandex-aistudio)
• Auth: YANDEX_API_KEY + YANDEX_FOLDER_ID
• OpenAI-compatible endpoint: https://llm.api.cloud.yandex.net/v1
• Sends x-folder-id and x-data-logging-enabled: false on requests
• Model ids normalized to Yandex gpt://<folder_id>/<model> URIs
• Does not send extra_body.thinking (rejected by the API for DeepSeek V4 Flash)
• Documents env vars in .env.example and provider in cli-config.yaml.example

Related

• Gateway ${VAR} expansion in some paths: [Bug]: Gateway environment variable not expanded for custom model providers path #21275, fix(gateway): expand config env templates before auth bridging and model resolution #35444
• No duplicate Yandex LLM provider PR found

Test plan

• pytest tests/hermes_cli/test_yandex_provider.py -q (13 tests)
• pytest tests/providers/test_plugin_discovery.py -q
• hermes doctor with real YANDEX_API_KEY + YANDEX_FOLDER_ID
• CLI chat with model.provider: yandex
• Gateway/Telegram after fix(gateway): expand config env templates before auth bridging and model resolution #35444 (or use literal gpt://<folder>/... in config until then)

Platform tested

• Linux (Ubuntu)

[tonydwb]

Code Review Summary

Verdict: Approved ✅

Review Findings

This PR adds the Yandex Cloud AI Studio as a new model provider — a new OpenAI-compatible provider plugin with full registry integration.

✅ Looks Good

• Clean provider profile: Follows the standard ProviderProfile pattern with all required fields (env_vars, auth_type, base_url, fallback_models).
• Model normalization: normalize_yandex_model() correctly produces gpt://<folder_id>/<model> URIs, with YANDEX_FOLDER_ID env var placeholder expansion support.
• Dynamic headers: YANDEX_FOLDER_ID is read at access time via _yandex_headers(), not cached at init — correct for runtime env changes.
• API extras: build_api_kwargs_extras() returns ({}, {}) — explicitly documents that Yandex rejects extra_body.thinking for DeepSeek V4 Flash. This prevents hard-to-debug errors.
• fetch_models: Uses stdlib urllib.request (no extra deps), proper auth header and timeout handling.
• Test coverage: 86 lines of tests covering registry registration, aliases, model normalization, build_api_kwargs_extras, and dynamic folder headers.
• Integration hooks: Updated .env.example, cli-config.yaml.example, and model_normalize.py.
• Plugin metadata: Proper plugin.yaml with standard fields.

No Issues Found

---

Reviewed by Hermes Agent

[tonydwb]

Code Review Summary

Verdict: Approved ✅

Review

Adds first-class Yandex Cloud AI Studio model provider — well-structured plugin with proper auth, model URI normalization, and thorough test coverage.

✅ Looks Good

• Complete plugin: provider id, aliases, auth (YANDEX_API_KEY + YANDEX_FOLDER_ID), OpenAI-compatible endpoint, model URI normalization.
• API compatibility: Sends x-folder-id and x-data-logging-enabled: false headers.
• Edge case handled: Strips extra_body.thinking which Yandex rejects.
• Good test coverage: 13 tests, plus .env.example and cli-config.yaml.example updated.
• Clean diff: 105 lines plugin + 86 lines tests + documentation updates.

---

Reviewed by Hermes Agent (cron job)

[teknium1]

Thanks for the complete Yandex profile, normalization logic, and focused tests. This is an automated hermes-sweeper review.

• The PR adds plugins/model-providers/yandex/ as a bundled model-provider (plugins/model-providers/yandex/plugin.yaml:1-2, plugins/model-providers/yandex/__init__.py:91-105).
• Current maintainer policy does not accept new third-party/vendor integrations in the core tree: AGENTS.md:126-136 requires them to ship as standalone plugin repositories because the project does not own or maintain those backends.
• The supported alternative is a standalone Yandex plugin installed at $HERMES_HOME/plugins/model-providers/yandex/ or distributed through a pip entry point; the existing provider discovery path supports both (website/docs/developer-guide/model-provider-plugin.md:9-23,249-258).

Please publish the provider as a standalone plugin repo and share it in #plugins-skills-and-skins; the policy is a coupling and maintenance decision, not a judgment on the implementation quality.

---

Closed as not-planned per standing maintainer policy (in-tree-provider-integration). This is a design-direction decision, not a code-quality judgment — see the Contribution Rubric in AGENTS.md for what the project is looking for. If you believe this policy was misapplied to your change, comment here and a maintainer will take a look.