Help on agents

[Bennoo]

Self Checks

• I have searched for existing issues search for existing issues, 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.

1. Is this request related to a challenge you're experiencing? Tell me about your story.

I have an agent node set as AgenticStrategy: FunctionCalling. The LLM set is a GPT5.1 through the openai API plugin, set as a llm with function call type: Function call.
The agent node have multiple tools available. But I cannot succeed in making it using the tools. How can I make it using tools?

2. Additional context or comments

No response

[dosubot[bot]]

The most common cause of agents not using tools is Maximum Iterations being set to 1. When it's set to 1, the tool list isn't sent to the LLM at all, so tools can't be called.

Quick fix: Open your agent node settings and ensure "Maximum Iterations" is set to at least 2 (recommended: 10).

If that doesn't resolve it, here's a troubleshooting checklist:

1. Verify tools are enabled — Check that all tools have their toggle turned on in the agent node
2. Check tool credentials — Ensure credentials haven't expired
3. Reduce tool count — Having too many tools (8+) can cause the LLM to skip using them
4. Check Dify version — Version 1.11.4+ includes fixes for tool invocation bugs
5. Add explicit system prompt — Try adding: "You MUST use the available tools to answer. Do NOT answer from your own knowledge."
6. Test with GPT-4o — This rules out any GPT-5.1 specific issues (though GPT-5.1 is properly configured for function calling in the plugin)

Let me know which step resolves your issue or if you need help debugging further!

To reply, just mention @dosu.

---

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

   

[dosubot[bot]]

Yes! You can enable debug logging to see the JSON payloads being sent to your gateway.

Add these to your .env file or docker-compose.yml:

LOG_LEVEL=DEBUG
ENABLE_REQUEST_LOGGING=True

Then restart the services:

docker-compose restart api plugin_daemon

To view the logs:

# Plugin daemon logs (these show the actual HTTP requests to your gateway)
docker-compose logs -f plugin_daemon

# API logs
docker-compose logs -f api

The plugin_daemon logs are the key ones for your use case — they'll show the actual HTTP requests going to your LLM gateway including the full JSON payload, headers, and response details.

With debug logging enabled, you'll see entries like:

• Request URLs and methods
• Request/response bodies in formatted JSON
• HTTP status codes and error details

This should help you identify what's causing the 400/500 errors from your gateway!

To reply, just mention @dosu.

---

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

   

[Bennoo]

@dosu are LOG_LEVEL=DEBUG
ENABLE_REQUEST_LOGGING=True for the plugin deamon or the worker?

[dosubot[bot]]

Both! These environment variables are shared across multiple services via a YAML anchor in docker-compose.yaml:

• api
• worker
• worker_beat
• plugin_daemon

So setting LOG_LEVEL=DEBUG and ENABLE_REQUEST_LOGGING=True in your .env file will enable debug logging for all of them, including the plugin_daemon where the actual LLM gateway requests are made.

After adding to your .env, restart all services:

docker-compose restart api worker plugin_daemon

Then check docker-compose logs -f plugin_daemon to see the requests going to your gateway.

To reply, just mention @dosu.

---

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

   

[Bennoo]

@dosu I set ENABLE_REQUEST_LOGGING=True and log level but I cannot see the JSON sent to LLM gateway

[dosubot[bot]]

I apologize for the misleading info earlier! After checking the code, ENABLE_REQUEST_LOGGING only logs incoming requests to Dify's API, not outgoing requests to LLM providers.

Unfortunately, Dify doesn't have built-in logging for the actual JSON payloads sent to LLM gateways.

To capture the actual requests to your gateway, you'll need to use network-level debugging:

1. mitmproxy (recommended) — Set it up as an HTTP proxy between Dify and your gateway:

   mitmproxy --mode reverse:https://your-gateway-url

   Then point Dify's OpenAI-compatible endpoint to the mitmproxy URL.

2. Gateway-side logging — If you control the gateway, enable request logging there to see what Dify is sending.

3. tcpdump/Wireshark — Capture network traffic directly on the container or host.

This should help you see exactly what's being sent and identify why you're getting the 400/500 errors.

To reply, just mention @dosu.

---

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