This is using gpt-oss:120b throught vLLM.
I enabled reasoning for an Agent. Sometimes I see it’s using a tool (Used create_reasoning_plan (1)) and sometimes not. How is this decision made? I looked at the documentation but couldn’t find anything related to that tool.
🚀 Crew: crew
├── 📋 Task: extractor_task (ID: 09b4162c-f6ee-4d54-a797-011cf285bd6f)
│ Assigned to: Device, interface and time extractor
│
│ Status: ✅ Completed
│ ├── ✅ Reasoning Completed
│ ├── 🔧 Used create_reasoning_plan (1)
│ └── ✅ Knowledge Retrieval Completed
├── ✅ Knowledge Retrieval Completed
└── 📋 Task: elasticsearch_lldp_task (ID: 7e21214b-756d-4a75-b026-9559f417b0ca)
Assigned to: You are an expert in Elasticsearch Query DSL.
Status: ✅ Completed
Other times (most of the time), I see the following:
🚀 Crew: crew
├── 📋 Task: extractor_task (ID: df19f2b6-b808-4709-bf39-0ec55bfff2ee)
│ Assigned to: Device, interface and time extractor
│
│ Status: ✅ Completed
│ ├── ✅ Reasoning Completed
│ └── ✅ Knowledge Retrieval Completed
├── ✅ Knowledge Retrieval Completed
└── 📋 Task: elasticsearch_lldp_task (ID: a01183b9-6c37-451d-b818-d7bec25a6523)
Assigned to: You are an expert in Elasticsearch Query DSL.
Status: ✅ Completed
In both cases, the reasoning plan looked good but the task failed to abide by the plan and/or hallucinate.