Got empty response intermittently from LLM using Agent Node #886

New Issue

Closed

opened 2026-02-16 10:20:51 -05:00 by yindo · 5 comments

yindo commented 2026-02-16 10:20:51 -05:00

Owner

Copy Link

Copy Source

Originally created by @Pandaaaa906 on GitHub (Dec 19, 2025).

Self Checks

• This is only for bug report, if you would like to ask a question, please head to Discussions.
• I have searched for existing issues Dify issues & Dify Official Plugins, including closed ones.
• I confirm that I am using English to submit this report (我已阅读并同意 Language Policy).
• [FOR CHINESE USERS] 请务必使用英文提交 Issue，否则会被关闭。谢谢！:）
• Please do not modify this template :) and fill in all the required fields.

Dify version

1.10.0

Plugin version

0.0.25(0.2.4 for mcp_see_agent)

Cloud or Self Hosted

Self Hosted (Docker)

Steps to reproduce

my flow setup like this, with ReAct strategy

it will get "I am thinking about how to help you" from time to time,

it's turn out that LLM response empty response

i modify the plugin to let it retry in next iteration when this happen,
https://github.com/Pandaaaa906/dify-plugin-agent-mcp_sse/tree/fix-retry-llm-false-format
the llm's response will be normal again,
but in the next iteration step it will empty response again....

Models i test:
volcengine: kimi-k2-thinking bad, in another level
volcengine: kimi-k2 always bad
volcengine: Qwen/Qwen2.5-32B always bad
siliconflow: moonshotai/Kimi-K2-Instruct good
siliconflow: deepseek-ai/DeekSeek-R1 bad
moonshotai: kimi-k2-thinking good but real slow

✔️ Error log

No response

Originally created by @Pandaaaa906 on GitHub (Dec 19, 2025). ### Self Checks - [x] This is only for bug report, if you would like to ask a question, please head to [Discussions](https://github.com/langgenius/dify/discussions/categories/general). - [x] I have searched for existing issues [Dify issues](https://github.com/langgenius/dify/issues) & [Dify Official Plugins](https://github.com/langgenius/dify-official-plugins/issues), including closed ones. - [x] I confirm that I am using English to submit this report (我已阅读并同意 [Language Policy](https://github.com/langgenius/dify/issues/1542)). - [x] [FOR CHINESE USERS] 请务必使用英文提交 Issue，否则会被关闭。谢谢！:） - [x] Please do not modify this template :) and fill in all the required fields. ### Dify version 1.10.0 ### Plugin version 0.0.25(0.2.4 for mcp_see_agent) ### Cloud or Self Hosted Self Hosted (Docker) ### Steps to reproduce my flow setup like this, with ReAct strategy <img width="1829" height="792" alt="Image" src="https://github.com/user-attachments/assets/160f8adf-07b3-4763-afc8-3b1e0767df37" /> it will get "I am thinking about how to help you" from time to time, it's turn out that LLM response empty response <img width="1325" height="315" alt="Image" src="https://github.com/user-attachments/assets/288f8c72-cd79-4c3e-ba5c-63aa7e91d064" /> i modify the plugin to let it retry in next iteration when this happen, https://github.com/Pandaaaa906/dify-plugin-agent-mcp_sse/tree/fix-retry-llm-false-format the llm's response will be normal again, but in the next iteration step it will empty response again.... Models i test: volcengine: kimi-k2-thinking bad, in another level volcengine: kimi-k2 always bad volcengine: Qwen/Qwen2.5-32B always bad siliconflow: moonshotai/Kimi-K2-Instruct good siliconflow: deepseek-ai/DeekSeek-R1 bad moonshotai: kimi-k2-thinking good but real slow ### ✔️ Error log _No response_

yindo added the bug label 2026-02-16 10:20:51 -05:00

yindo closed this issue 2026-02-16 10:20:51 -05:00

yindo commented 2026-02-16 10:20:52 -05:00

Author

Owner

Copy Link

Copy Source

@dosubot[bot] commented on GitHub (Dec 19, 2025):

This is a known issue in Dify when using Agent Node with the ReAct strategy—intermittent empty LLM responses (often surfacing as "I am thinking about how to help you") have been reported by multiple users across versions, especially with certain models and in flows involving tool calls or loops. The root causes include agent strategy failures, tool/plugin output quirks, model API inconsistencies, and lack of defensive checks or retries in the core agent and ReAct code. Some models and plugins (like Qwen/Tongyi or kimi-k2) are more prone to this due to schema or output format issues, while others (like moonshotai/Kimi-K2-Instruct) perform better in this context example.

There is no universal fix yet, but these steps can help reduce the frequency or impact:

• Limit the number of tools per agent node (ideally <4), and double-check input/output configs for all nodes reference.
• Use models and plugin versions known to be more stable with ReAct (e.g., moonshotai/Kimi-K2-Instruct, certain Tongyi plugin versions) reference.
• Add retry logic for LLM invocations in your plugin or flow (with exponential backoff), as you’ve started—this is now a best practice in several official plugins reference.
• Tune error handling modes in iteration/loop nodes (e.g., CONTINUE_ON_ERROR) to avoid halting on empty outputs reference.
• Increase buffer and workflow limits in your environment, though this alone rarely solves the issue reference.
• For Qwen/Tongyi, ensure the plugin manifest includes all required fields (like description), and try plugin versions 0.0.18 or 0.0.7 if 0.0.25 is problematic reference.

The core Dify Agent Node and ReAct strategy currently do not retry or filter empty LLM responses—they surface them as-is, so plugin-level workarounds (like your retry patch) are the most effective mitigation for now reference.

If you want to debug further, enable detailed logs, isolate which models/tools trigger the issue most, and try reducing flow complexity. If you can consistently reproduce the problem with a specific model or plugin version, sharing those details (and any error logs) will help maintainers prioritize a more robust fix.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

  

@dosubot[bot] commented on GitHub (Dec 19, 2025): <!-- Answer --> This is a known issue in Dify when using Agent Node with the ReAct strategy—intermittent empty LLM responses (often surfacing as "I am thinking about how to help you") have been reported by multiple users across versions, especially with certain models and in flows involving tool calls or loops. The root causes include agent strategy failures, tool/plugin output quirks, model API inconsistencies, and lack of defensive checks or retries in the core agent and ReAct code. Some models and plugins (like Qwen/Tongyi or kimi-k2) are more prone to this due to schema or output format issues, while others (like moonshotai/Kimi-K2-Instruct) perform better in this context [example](https://github.com/langgenius/dify-official-plugins/issues/1704). There is no universal fix yet, but these steps can help reduce the frequency or impact: - Limit the number of tools per agent node (ideally <4), and double-check input/output configs for all nodes [reference](https://github.com/langgenius/dify/issues/26464). - Use models and plugin versions known to be more stable with ReAct (e.g., moonshotai/Kimi-K2-Instruct, certain Tongyi plugin versions) [reference](https://github.com/langgenius/dify-official-plugins/issues/1258). - Add retry logic for LLM invocations in your plugin or flow (with exponential backoff), as you’ve started—this is now a best practice in several official plugins [reference](https://github.com/langgenius/dify-official-plugins/pull/2065). - Tune error handling modes in iteration/loop nodes (e.g., CONTINUE_ON_ERROR) to avoid halting on empty outputs [reference](https://github.com/langgenius/dify/issues/26464). - Increase buffer and workflow limits in your environment, though this alone rarely solves the issue [reference](https://github.com/langgenius/dify/issues/26464). - For Qwen/Tongyi, ensure the plugin manifest includes all required fields (like description), and try plugin versions 0.0.18 or 0.0.7 if 0.0.25 is problematic [reference](https://github.com/langgenius/dify-official-plugins/issues/1258). The core Dify Agent Node and ReAct strategy currently do not retry or filter empty LLM responses—they surface them as-is, so plugin-level workarounds (like your retry patch) are the most effective mitigation for now [reference](https://github.com/langgenius/dify-official-plugins/issues/1704). If you want to debug further, enable detailed logs, isolate which models/tools trigger the issue most, and try reducing flow complexity. If you can consistently reproduce the problem with a specific model or plugin version, sharing those details (and any error logs) will help maintainers prioritize a more robust fix. <!-- Dosu Comment Footer --> *To reply, just mention [@dosu](https://go.dosu.dev/dosubot).* --- ^{How did I do? [Good](https://app.dosu.dev/response-feedback/46bd50a4-72dc-42ac-8e90-4543379374e3?feedback_type=great_response) | [Irrelevant](https://app.dosu.dev/response-feedback/46bd50a4-72dc-42ac-8e90-4543379374e3?feedback_type=irrelevant_answer) | [Incorrect](https://app.dosu.dev/response-feedback/46bd50a4-72dc-42ac-8e90-4543379374e3?feedback_type=incorrect_sources) | [Verbose](https://app.dosu.dev/response-feedback/46bd50a4-72dc-42ac-8e90-4543379374e3?feedback_type=too_verbose) | [Hallucination](https://app.dosu.dev/response-feedback/46bd50a4-72dc-42ac-8e90-4543379374e3?feedback_type=hallucination) | [Report 🐛](https://app.dosu.dev/response-feedback/46bd50a4-72dc-42ac-8e90-4543379374e3?feedback_type=bug_report) | [Other](https://app.dosu.dev/response-feedback/46bd50a4-72dc-42ac-8e90-4543379374e3?feedback_type=other)} [](https://cursor.com/link/prompt?text=This%20is%20a%20known%20issue%20in%20Dify%20when%20using%20Agent%20Node%20with%20the%20ReAct%20strategy%E2%80%94intermittent%20empty%20LLM%20responses%20%28often%20surfacing%20as%20%22I%20am%20thinking%20about%20how%20to%20help%20you%22%29%20have%20been%20reported%20by%20multiple%20users%20across%20versions%2C%20especially%20with%20certain%20models%20and%20in%20flows%20involving%20tool%20calls%20or%20loops.%20The%20root%20causes%20include%20agent%20strategy%20failures%2C%20tool/plugin%20output%20quirks%2C%20model%20API%20inconsistencies%2C%20and%20lack%20of%20defensive%20checks%20or%20retries%20in%20the%20core%20agent%20and%20ReAct%20code.%20Some%20models%20and%20plugins%20%28like%20Qwen/Tongyi%20or%20kimi-k2%29%20are%20more%20prone%20to%20this%20due%20to%20schema%20or%20output%20format%20issues%2C%20while%20others%20%28like%20moonshotai/Kimi-K2-Instruct%29%20perform%20better%20in%20this%20context%C2%A0%5Bexample%5D%28https%3A//github.com/langgenius/dify-official-plugins/issues/1704%29.%0A%0AThere%20is%20no%20universal%20fix%20yet%2C%20but%20these%20steps%20can%20help%20reduce%20the%20frequency%20or%20impact%3A%0A%0A-%20Limit%20the%20number%20of%20tools%20per%20agent%20node%20%28ideally%20%3C4%29%2C%20and%20double-check%20input/output%20configs%20for%20all%20nodes%C2%A0%5Breference%5D%28https%3A//github.com/langgenius/dify/issues/26464%29.%0A-%20Use%20models%20and%20plugin%20versions%20known%20to%20be%20more%20stable%20with%20ReAct%20%28e.g.%2C%20moonshotai/Kimi-K2-Instruct%2C%20certain%20Tongyi%20plugin%20versions%29%C2%A0%5Breference%5D%28https%3A//github.com/langgenius/dify-official-plugins/issues/1258%29.%0A-%20Add%20retry%20logic%20for%20LLM%20invocations%20in%20your%20plugin%20or%20flow%20%28with%20exponential%20backoff%29%2C%20as%20you%E2%80%99ve%20started%E2%80%94this%20is%20now%20a%20best%20practice%20in%20several%20official%20plugins%C2%A0%5Breference%5D%28https%3A//github.com/langgenius/dify-official-plugins/pull/2065%29.%0A-%20Tune%20error%20handling%20modes%20in%20iteration/loop%20nodes%20%28e.g.%2C%20CONTINUE_ON_ERROR%29%20to%20avoid%20halting%20on%20empty%20outputs%C2%A0%5Breference%5D%28https%3A//github.com/langgenius/dify/issues/26464%29.%0A-%20Increase%20buffer%20and%20workflow%20limits%20in%20your%20environment%2C%20though%20this%20alone%20rarely%20solves%20the%20issue%C2%A0%5Breference%5D%28https%3A//github.com/langgenius/dify/issues/26464%29.%0A-%20For%20Qwen/Tongyi%2C%20ensure%20the%20plugin%20manifest%20includes%20all%20required%20fields%20%28like%20description%29%2C%20and%20try%20plugin%20versions%200.0.18%20or%200.0.7%20if%200.0.25%20is%20problematic%C2%A0%5Breference%5D%28https%3A//github.com/langgenius/dify-official-plugins/issues/1258%29.%0A%0AThe%20core%20Dify%20Agent%20Node%20and%20ReAct%20strategy%20currently%20do%20not%20retry%20or%20filter%20empty%20LLM%20responses%E2%80%94they%20surface%20them%20as-is%2C%20so%20plugin-level%20workarounds%20%28like%20your%20retry%20patch%29%20are%20the%20most%20effective%20mitigation%20for%20now%C2%A0%5Breference%5D%28https%3A//github.com/langgenius/dify-official-plugins/issues/1704%29.%0A%0AIf%20you%20want%20to%20debug%20further%2C%20enable%20detailed%20logs%2C%20isolate%20which%20models/tools%20trigger%20the%20issue%20most%2C%20and%20try%20reducing%20flow%20complexity.%20If%20you%20can%20consistently%20reproduce%20the%20problem%20with%20a%20specific%20model%20or%20plugin%20version%2C%20sharing%20those%20details%20%28and%20any%20error%20logs%29%20will%20help%20maintainers%20prioritize%20a%20more%20robust%20fix.)&nbsp;[](https://go.dosu.dev/discord-bot)&nbsp;[](https://twitter.com/intent/tweet?text=%40dosu_ai%20helped%20me%20solve%20this%20issue!&url=https%3A//github.com/langgenius/dify-official-plugins/issues/2295)

yindo commented 2026-02-16 10:20:52 -05:00

Author

Owner

Copy Link

Copy Source

@Pandaaaa906 commented on GitHub (Dec 19, 2025):

here r some related issues
https://github.com/langgenius/dify/issues/20499
https://github.com/langgenius/dify/issues/26464

@Pandaaaa906 commented on GitHub (Dec 19, 2025): here r some related issues https://github.com/langgenius/dify/issues/20499 https://github.com/langgenius/dify/issues/26464

yindo commented 2026-02-16 10:20:52 -05:00

Author

Owner

Copy Link

Copy Source

@pr-maia commented on GitHub (Dec 19, 2025):

It'd be great to have this fixed or gain the ability to configure retries for the agent ndoe

@pr-maia commented on GitHub (Dec 19, 2025): It'd be great to have this fixed or gain the ability to configure retries for the agent ndoe

yindo commented 2026-02-16 10:20:52 -05:00

Author

Owner

Copy Link

Copy Source

@Pandaaaa906 commented on GitHub (Dec 19, 2025):

@pr-maia i've created a pull request that could retry when LLM got empty response or false format in the unofficial agent plugin

@Pandaaaa906 commented on GitHub (Dec 19, 2025): @pr-maia i've created a pull request that could retry when LLM got empty response or false format in the unofficial agent plugin

yindo commented 2026-02-16 10:20:52 -05:00

Author

Owner

Copy Link

Copy Source

@dosubot[bot] commented on GitHub (Jan 4, 2026):

Hi, @Pandaaaa906. I'm Dosu, and I'm helping the dify-official-plugins team manage their backlog and am marking this issue as stale.

Issue Summary:

• You reported intermittent empty responses from LLMs using the Agent Node with the ReAct strategy in Dify v1.10.0, especially with certain models like volcengine's kimi-k2 and Qwen.
• The problem is linked to agent strategy failures, inconsistencies in model APIs, and the lack of retry logic in the core code.
• Workarounds include limiting the number of tools per node, using more stable models, adding retry logic, and tuning error handling.
• You contributed a pull request adding retry logic for empty or malformed LLM responses in an unofficial agent plugin.
• Another user showed interest in making retries configurable for the agent node, showing ongoing progress.

Next Steps:

• Please let me know if this issue is still relevant with the latest version of dify-official-plugins by commenting here to keep the discussion open.
• Otherwise, this issue will be automatically closed in 5 days.

Thank you for your understanding and valuable contribution!

@dosubot[bot] commented on GitHub (Jan 4, 2026): Hi, @Pandaaaa906. I'm [Dosu](https://dosu.dev), and I'm helping the dify-official-plugins team manage their backlog and am marking this issue as stale. **Issue Summary:** - You reported intermittent empty responses from LLMs using the Agent Node with the ReAct strategy in Dify v1.10.0, especially with certain models like volcengine's kimi-k2 and Qwen. - The problem is linked to agent strategy failures, inconsistencies in model APIs, and the lack of retry logic in the core code. - Workarounds include limiting the number of tools per node, using more stable models, adding retry logic, and tuning error handling. - You contributed a pull request adding retry logic for empty or malformed LLM responses in an unofficial agent plugin. - Another user showed interest in making retries configurable for the agent node, showing ongoing progress. **Next Steps:** - Please let me know if this issue is still relevant with the latest version of dify-official-plugins by commenting here to keep the discussion open. - Otherwise, this issue will be automatically closed in 5 days. Thank you for your understanding and valuable contribution!

Sign in to join this conversation.

No Branch/Tag Specified

Branches Tags

main

dependabot/uv/datasources/google_cloud_storage/uv-0225d4cafc

dependabot/uv/tools/chart/uv-cdf8d38825

feat/moonshot-kimi-k3-k27-code

feat/gemini-system-instruction-semantics

fix/gemini-text-system-parts

dosu/doc-update-3436

feat/openai-gpt-5-6-support

fix/outlook-trigger-readme-permissions

fix/microsoft-oauth-datasource-tenant

fix/byteplus-api-base-url-label

feat/ark-video-generation-custom-models

fix/openai-stream-reasoning-thinking

fix/skip-live-tests-missing-keys

codex/enhance-discord-integrations

codex/harden-trigger-oauth-permissions

update-lemonade-default-port

codex/fix-pr-review-ci

feat/ark-multimodal-polling

copilot/update-mineru-3x-plugin-file-types

feat/action-security-fix

feat/claude-fable5

codex/fix-ollama-model-category

codex/upgrade-ollama-latest

feat/minimax-m3

codex/refine-plugin-readmes

fix/fireworks-model-label-zh-hans

chore/version-bump-plugins-batch-06

chore/version-bump-plugins-batch-05

chore/version-bump-plugins-batch-04

chore/version-bump-plugins-batch-03

chore/version-bump-plugins-batch-02

fix/plugin-sdk-i18n-fields

chore/version-bump-plugins-batch-01

chore/dependencies-high-priority

chore/dependencies-medium-priority

chore/dependencies-low-priority-clean

chore/dependencies-low-priority

fix/tongyi-align-deepseek-model-metadata

codex/update-openrouter-models

codex/azure-ai-studio-cjk-version-bump

feat/ernie-image-tool

claude/cranky-thompson-b6a281

fix/remove-unavailable-volcengine-models

fix/legacy-plugin-daemon-compatibility

copilot/add-deepseek-v4-to-openrouter

codex/merge-tongyi-tts-prs

codex/depot-runners

chore/migrate-selected-plugins-to-uv-lock

fix/repair-failed-plugin-deploys

ci/fix-plugin-change-detection

chore/migrate-aws-s3-storage-to-uv-lock

chore/migrate-somark-dependencies-to-uv-lock

ci/uv-lock-dependency-validation

fix/2911-openai-api-compatible-tool-call-content

chore/remove-deprecated-plugins

copilot/add-more-models-openrouter-provider

copilot/fix-error-connecting-fireworks-ai

fix/openai-compatible-max-completion-tokens-validation

fix/gemini-lockfile

chore/add-React-fallback-desc

feat/add-gemini-embedding-2-preview

copilot/add-claude-sonnet-opus-4-6

codex/update-xai-models-2721

fix/remove-sentence-transformer-for-hubspot

feat/add-gpt-5.4

copilot/fix-ci-errors-and-bump-version

copilot/fix-security-errors-combined

copilot/update-gemini-model-version

fix/anthropic-parallel-tool-call

codex/fix-azure-ai-studio-azure-core-1-38

codex/fix-azure-blob-azure-core-1-38

codex/fix-azure-openai-azure-core-1-38

gitlab-credentials-c

fix/openrouter-validation-model

fix/gemini-null-checks

fix/deepseek-tool-message-ignore

fix/gemini-dependencies

No results found.

Labels

Clear labels

bug

bug

bug

dependencies

documentation

enhancement

good first issue

invalid

pull-request

Mirrored from GitHub Pull Request

question

stale

No Label bug

Milestone

No items

No Milestone

Projects

Clear projects

No project

Assignees

Clear assignees

yindo

No Assignees

1 Participants

Notifications

Subscribe

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: langgenius/dify-official-plugins#886