fix: deep research structured output and recursion limit

[xinquiry]

Summary

• Use method="function_calling" for structured output in deep research clarify node
• Increase agent recursion limit from 25 to 50

Problem

• The default json_mode for with_structured_output() doesn't work with Claude models via GPUGEEK provider
• Claude supports tool/function calling natively but not OpenAI's response_format JSON mode
• Deep research agent was hitting recursion limit (25) on complex tasks

Solution

• Use method="function_calling" which works across all providers (OpenAI, Gemini, Claude)
• Increase recursion_limit to 50 in agent.astream() config

Test plan

• Tested with ultra tier (Claude via GPUGEEK) - structured output works
• Test with other tiers to verify no regression

🤖 Generated with Claude Code

Summary by Sourcery

调整深度研究澄清流程，以使用与提供商无关的结构化输出，并提升智能体在复杂任务中的递归能力。

Bug 修复：

• 将深度研究澄清节点的结构化输出切换为使用函数调用，以便在不同的 LLM 提供商之间保持兼容性。

增强功能：

• 将流式配置中的智能体递归限制从 25 提升到 50，以更好地处理复杂的深度研究工作流。

Original summary in English

Summary by Sourcery

Adjust deep research clarification to use provider-agnostic structured output and increase agent recursion capacity for complex tasks.

Bug Fixes:

• Switch deep research clarify node structured output to use function calling for compatibility across LLM providers.

Enhancements:

• Raise the agent recursion limit from 25 to 50 in the streaming config to better handle complex deep research workflows.

[sourcery-ai]

审阅者指南（在小型 PR 上折叠）

审阅者指南

调整 deep research clarify 节点以使用基于 function calling 的结构化输出，从而获得更好的多模型兼容性，并提升 agent 的递归上限，使其在终止前可以处理更复杂的任务。

使用 function_calling 的 clarify_node 结构化输出时序图

sequenceDiagram
    participant ClarifyNode
    participant LLMFactory
    participant LLM
    participant StructuredLLM

    ClarifyNode->>LLMFactory: llm_factory(temperature=0.3)
    LLMFactory-->>ClarifyNode: llm

    ClarifyNode->>ClarifyNode: messages_str = get_buffer_string(state.messages)
    ClarifyNode->>ClarifyNode: date_str = get_today_str()
    ClarifyNode->>ClarifyNode: prompt = CLARIFY_WITH_USER_PROMPT.format(messages, date)

    ClarifyNode->>LLM: with_structured_output(ClarifyWithUser, method=function_calling)
    LLM-->>ClarifyNode: llm_with_struct

    ClarifyNode->>StructuredLLM: ainvoke([HumanMessage(content=prompt)])
    StructuredLLM-->>ClarifyNode: response

    ClarifyNode->>ClarifyNode: result = ClarifyWithUser.model_validate(response)
    ClarifyNode-->>ClarifyNode: use result for clarification state

Loading

提升 recursion_limit 后的 agent.astream 时序图

sequenceDiagram
    participant ChatProcessor
    participant Agent

    ChatProcessor->>ChatProcessor: prepare history_messages

    ChatProcessor->>Agent: astream({messages: history_messages}, stream_mode=[updates, messages], config={recursion_limit: 50})
    loop for each chunk
        Agent-->>ChatProcessor: (mode, data)
        ChatProcessor->>ChatProcessor: chunk_count += 1
        ChatProcessor->>ChatProcessor: handle mode and data
    end

    Agent-->>ChatProcessor: stream completed

Loading

deep research clarify 与 agent 流式变更的类图

classDiagram
    class ClarifyState {
    }

    class ClarifyWithUser {
    }

    class LLMFactory {
        +llm_factory(temperature)
    }

    class LLM {
        +with_structured_output(output_type, method)
    }

    class StructuredLLM {
        +ainvoke(messages)
    }

    class ClarifyNode {
        +clarify_node(state)
    }

    class Agent {
        +astream(input, stream_mode, config)
    }

    class ChatProcessor {
        +_process_agent_stream(agent, history_messages)
    }

    ClarifyNode --> ClarifyState
    ClarifyNode --> ClarifyWithUser
    ClarifyNode --> LLMFactory
    LLMFactory --> LLM
    LLM --> StructuredLLM
    ClarifyNode --> StructuredLLM

    ChatProcessor --> Agent
    ChatProcessor --> ClarifyNode
    Agent ..> ClarifyWithUser
    Agent ..> ClarifyState
    StructuredLLM ..> ClarifyWithUser
    ClarifyWithUser <.. ChatProcessor

Loading

文件级变更

变更	详情	文件
将 clarify 节点的结构化输出切换为使用 function-calling，以实现跨模型提供商的兼容性。	与之前一样，通过现有的 llm_factory 并使用 temperature 0.3 获取基础 LLM 从缓冲消息和当前日期构建 clarify prompt，无需在单独步骤中将 messages 转换为列表 使用 ClarifyWithUser 和 method="function_calling" 调用 with_structured_output 包装 LLM 使用人类消息 prompt 异步调用结构化 LLM，并在需要时通过 model_validate 将响应规范化为 ClarifyWithUser	service/app/agents/components/deep_research/components.py
提升 agent 的递归上限，以便在流式运行中支持更深层次推理。	更新 _process_agent_stream，在调用 agent.astream 时传入包含 recursion_limit=50 的 config 字典 在新增递归上限的同时，保留现有的 stream_mode 选项 ["updates", "messages"] 及其它行为	service/app/core/chat/langchain.py

---

提示与命令

与 Sourcery 交互

• 触发一次新的审阅： 在 pull request 上评论 @sourcery-ai review。
• 继续讨论： 直接回复 Sourcery 的审阅评论。
• 从审阅评论生成 GitHub issue： 通过回复某条审阅评论，请求 Sourcery 根据该评论创建 issue。你也可以在审阅评论下回复 @sourcery-ai issue，从该评论创建一个 issue。
• 生成 pull request 标题： 在 pull request 标题的任意位置写上 @sourcery-ai，即可在任意时间生成标题。你也可以在 pull request 上评论 @sourcery-ai title 来（重新）生成标题。
• 生成 pull request 摘要： 在 pull request 正文任意位置写上 @sourcery-ai summary，即可在你想要的位置随时生成 PR 摘要。你也可以在 pull request 上评论 @sourcery-ai summary 来（重新）生成摘要。
• 生成审阅者指南： 在 pull request 上评论 @sourcery-ai guide，即可在任意时间（重新）生成审阅者指南。
• 解决所有 Sourcery 评论： 在 pull request 上评论 @sourcery-ai resolve，即可标记解决所有 Sourcery 评论。如果你已经处理完所有评论且不想再看到它们，这会很有用。
• 忽略所有 Sourcery 审阅： 在 pull request 上评论 @sourcery-ai dismiss，即可忽略所有现有的 Sourcery 审阅。特别适用于你想从头开始新的审阅时——别忘了再评论 @sourcery-ai review 来触发新的审阅！

自定义你的体验

访问你的 控制面板 以：

• 启用或禁用审阅特性，如 Sourcery 生成的 pull request 摘要、审阅者指南等。
• 更改审阅语言。
• 添加、移除或编辑自定义审阅指令。
• 调整其它审阅设置。

获取帮助

• 如有问题或反馈，请联系支持团队。
• 访问我们的文档以获取详细指南和信息。
• 通过关注我们的 X/Twitter、LinkedIn 或 GitHub 与 Sourcery 团队保持联系。

Original review guide in English

Reviewer's guide (collapsed on small PRs)

Reviewer's Guide

Adjusts the deep research clarify node to use function-calling-based structured output for better provider compatibility and raises the agent recursion limit to handle more complex tasks before terminating.

Sequence diagram for clarify_node structured output using function_calling

sequenceDiagram
    participant ClarifyNode
    participant LLMFactory
    participant LLM
    participant StructuredLLM

    ClarifyNode->>LLMFactory: llm_factory(temperature=0.3)
    LLMFactory-->>ClarifyNode: llm

    ClarifyNode->>ClarifyNode: messages_str = get_buffer_string(state.messages)
    ClarifyNode->>ClarifyNode: date_str = get_today_str()
    ClarifyNode->>ClarifyNode: prompt = CLARIFY_WITH_USER_PROMPT.format(messages, date)

    ClarifyNode->>LLM: with_structured_output(ClarifyWithUser, method=function_calling)
    LLM-->>ClarifyNode: llm_with_struct

    ClarifyNode->>StructuredLLM: ainvoke([HumanMessage(content=prompt)])
    StructuredLLM-->>ClarifyNode: response

    ClarifyNode->>ClarifyNode: result = ClarifyWithUser.model_validate(response)
    ClarifyNode-->>ClarifyNode: use result for clarification state

Loading

Sequence diagram for agent.astream with increased recursion_limit

sequenceDiagram
    participant ChatProcessor
    participant Agent

    ChatProcessor->>ChatProcessor: prepare history_messages

    ChatProcessor->>Agent: astream({messages: history_messages}, stream_mode=[updates, messages], config={recursion_limit: 50})
    loop for each chunk
        Agent-->>ChatProcessor: (mode, data)
        ChatProcessor->>ChatProcessor: chunk_count += 1
        ChatProcessor->>ChatProcessor: handle mode and data
    end

    Agent-->>ChatProcessor: stream completed

Loading

Class diagram for deep research clarify and agent streaming changes

classDiagram
    class ClarifyState {
    }

    class ClarifyWithUser {
    }

    class LLMFactory {
        +llm_factory(temperature)
    }

    class LLM {
        +with_structured_output(output_type, method)
    }

    class StructuredLLM {
        +ainvoke(messages)
    }

    class ClarifyNode {
        +clarify_node(state)
    }

    class Agent {
        +astream(input, stream_mode, config)
    }

    class ChatProcessor {
        +_process_agent_stream(agent, history_messages)
    }

    ClarifyNode --> ClarifyState
    ClarifyNode --> ClarifyWithUser
    ClarifyNode --> LLMFactory
    LLMFactory --> LLM
    LLM --> StructuredLLM
    ClarifyNode --> StructuredLLM

    ChatProcessor --> Agent
    ChatProcessor --> ClarifyNode
    Agent ..> ClarifyWithUser
    Agent ..> ClarifyState
    StructuredLLM ..> ClarifyWithUser
    ClarifyWithUser <.. ChatProcessor

Loading

File-Level Changes

Change	Details	Files
Switch clarify node structured output to use function-calling for cross-provider compatibility.	Obtain base LLM via existing llm_factory with temperature 0.3 as before Build the clarify prompt from buffered messages and current date without casting messages to list in a separate step Wrap the LLM with with_structured_output using ClarifyWithUser and method="function_calling" Invoke the structured LLM asynchronously with the human message prompt and normalize the response to ClarifyWithUser via model_validate when needed	service/app/agents/components/deep_research/components.py
Increase agent recursion limit to allow deeper reasoning in streamed agent runs.	Update _process_agent_stream to pass a config dict with recursion_limit set to 50 into agent.astream Preserve existing stream_mode options ["updates", "messages"] and other behavior while adding the new limit	service/app/core/chat/langchain.py

---

Tips and commands

Interacting with Sourcery

• Trigger a new review: Comment @sourcery-ai review on the pull request.
• Continue discussions: Reply directly to Sourcery's review comments.
• Generate a GitHub issue from a review comment: Ask Sourcery to create an
  issue from a review comment by replying to it. You can also reply to a
  review comment with @sourcery-ai issue to create an issue from it.
• Generate a pull request title: Write @sourcery-ai anywhere in the pull
  request title to generate a title at any time. You can also comment
  @sourcery-ai title on the pull request to (re-)generate the title at any time.
• Generate a pull request summary: Write @sourcery-ai summary anywhere in
  the pull request body to generate a PR summary at any time exactly where you
  want it. You can also comment @sourcery-ai summary on the pull request to
  (re-)generate the summary at any time.
• Generate reviewer's guide: Comment @sourcery-ai guide on the pull
  request to (re-)generate the reviewer's guide at any time.
• Resolve all Sourcery comments: Comment @sourcery-ai resolve on the
  pull request to resolve all Sourcery comments. Useful if you've already
  addressed all the comments and don't want to see them anymore.
• Dismiss all Sourcery reviews: Comment @sourcery-ai dismiss on the pull
  request to dismiss all existing Sourcery reviews. Especially useful if you
  want to start fresh with a new review - don't forget to comment
  @sourcery-ai review to trigger a new review!

Customizing Your Experience

Access your dashboard to:

• Enable or disable review features such as the Sourcery-generated pull request
  summary, the reviewer's guide, and others.
• Change the review language.
• Add, remove or edit custom review instructions.
• Adjust other review settings.

Getting Help

• Contact our support team for questions or feedback.
• Visit our documentation for detailed guides and information.
• Keep in touch with the Sourcery team by following us on X/Twitter, LinkedIn or GitHub.

[sourcery-ai]

Hey - 我在这里给出了一些整体性的反馈：

• _process_agent_stream 中硬编码的 recursion_limit=50 也许更适合作为可配置的设置或共享常量，这样就可以在不同环境中进行调优，而不需要修改代码。
• 在 clarify_node 中，如果这个模式在其他地方也会复用，建议对每个 LLM 实例（或在更高层级）只实例化一次 llm.with_structured_output(ClarifyWithUser, method="function_calling")，以避免分散的配置，并将与提供商相关的行为集中管理。

AI Agents 的提示词

Please address the comments from this code review:

## Overall Comments
- The hard-coded `recursion_limit=50` in `_process_agent_stream` may be better expressed as a configurable setting or shared constant so it can be tuned per environment without code changes.
- In `clarify_node`, consider instantiating `llm.with_structured_output(ClarifyWithUser, method="function_calling")` once per LLM instance (or at a higher level) if this pattern is reused elsewhere, to avoid scattered configuration and keep provider-specific behavior centralized.

---

Sourcery 对开源项目是免费的——如果你觉得我们的评审有帮助，欢迎分享 ✨

• X
• Mastodon
• LinkedIn
• Facebook

_{帮我变得更有用！请在每条评论上点击 👍 或 👎，我会根据你的反馈改进后续评审。}

Original comment in English

Hey - I've left some high level feedback:

• The hard-coded recursion_limit=50 in _process_agent_stream may be better expressed as a configurable setting or shared constant so it can be tuned per environment without code changes.
• In clarify_node, consider instantiating llm.with_structured_output(ClarifyWithUser, method="function_calling") once per LLM instance (or at a higher level) if this pattern is reused elsewhere, to avoid scattered configuration and keep provider-specific behavior centralized.

Prompt for AI Agents

Please address the comments from this code review:

## Overall Comments
- The hard-coded `recursion_limit=50` in `_process_agent_stream` may be better expressed as a configurable setting or shared constant so it can be tuned per environment without code changes.
- In `clarify_node`, consider instantiating `llm.with_structured_output(ClarifyWithUser, method="function_calling")` once per LLM instance (or at a higher level) if this pattern is reused elsewhere, to avoid scattered configuration and keep provider-specific behavior centralized.

---

Sourcery is free for open source - if you like our reviews please consider sharing them ✨

• X
• Mastodon
• LinkedIn
• Facebook

_{Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.}