[Fix] filter tool call capability for non-LLM models

[Procyon-Nan]

Summary

• Filter ToolCall from embedding and reranker model capabilities in BasePlatformClient.getModels().
• Keep the existing ImageGeneration handling and TextInput capability normalization unchanged.
• Ensure adapter-provided metadata cannot expose non-LLM models as tool-call capable through the platform registry.

Motivation

Some adapters can return embedding or reranker models with ModelCapabilities.ToolCall. Since model metadata is exposed through listAllModels() and listPlatformModels(), downstream UIs and plugins may incorrectly treat non-LLM models as supporting tool calls.

This fixes the issue at the core registry boundary instead of relying on UI-side filtering.

Validation

• yarn fast-build core
• yarn lint

[gemini-code-assist]

Code Review

This pull request updates the BasePlatformClient to filter out ToolCall capabilities for models of type embeddings or reranker. A review comment suggests adding a nullish coalescing guard to model.capabilities to prevent potential TypeError exceptions if the capabilities array is undefined or null.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: 2a015353-13ad-4a81-ac57-9a5ca5f96f78

📥 Commits

Reviewing files that changed from the base of the PR and between 7c4c1ea and dba67c0.

📒 Files selected for processing (2)

• packages/adapter-doubao/src/client.ts
• packages/adapter-openai/src/client.ts

🚧 Files skipped from review as they are similar to previous changes (2)

• packages/adapter-doubao/src/client.ts
• packages/adapter-openai/src/client.ts

---

Walkthrough

六个适配器包（Deepseek、Doubao、Hunyuan、OpenAI、OpenAI-like、Qwen）在 refreshModels() 中显式推导模型类型（llm/embeddings/reranker），并仅在类型为 LLM 时赋予 ModelCapabilities.ToolCall，非 LLM 类型不再默认包含该能力。

Changes

类型感知的模型能力过滤

Layer / File(s)	Summary
单源适配器：类型感知能力过滤 packages/adapter-deepseek/src/client.ts, packages/adapter-doubao/src/client.ts, packages/adapter-hunyuan/src/client.ts, packages/adapter-openai/src/client.ts	Deepseek、Doubao、Hunyuan 和 OpenAI 适配器在 refreshModels() 中显式推导 type（LLM 或 embeddings），并在构造 capabilities 时仅当 type === ModelType.llm 时才包含 ModelCapabilities.ToolCall；非 LLM 模型的能力列表为空或不包含 ToolCall。
OpenAI-like 适配器：additionalModels 和 rawModels 的双源过滤 packages/adapter-openai-like/src/client.ts	additionalModels 映射新增显式的 type 分支（embeddings/reranker/llm），并根据 type 调整 capabilities；对拉取到的 rawModels 的格式化改为仅当类型为 llm 时注入 supportToolCalling() 的 capabilities，否则置 capabilities: []。
Qwen 适配器：additionalModels 与 expandedModels 的双源过滤 packages/adapter-qwen/src/client.ts	additionalModels 按 modelType 推导 type 并在非 llm 时过滤移除 ModelCapabilities.ToolCall；expandedModels 的 type 与 capabilities 构造被重写为按类型条件包含 ToolCall 与 ImageInput。

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related PRs

• ChatLunaLab/chatluna#865: 同样修改 refreshModels() 中模型能力构造逻辑，涉及在适配器里注入/过滤特定能力（本次 PR 条件化 ToolCall，[Feature] support and harden native multimodal file handling #865 涉及 AudioInput 的注入与过滤）。

Suggested reviewers

• dingyi222666

Poem

🐰 模型各有其能力真，
大言未必能呼唤工具，
嵌入静默不需调用，
类型分明少了误判，
六处适配齐步看春风。

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	标题准确反映了 PR 的主要变更：过滤非 LLM 模型（嵌入式和重排名器）的工具调用能力。
Description check	✅ Passed	描述清晰阐述了 PR 的目的、动机和验证步骤，与变更集直接相关。
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
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.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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