How to properly parse a chat response when using Ollama provider?

[robiningelbrecht]

Ok, I feel really stupid. I'm using the Ollama provider with model llama3.2. When I call

$response = $this->agent->chat(new UserMessage($userInput));

$response->getContent() contains following info:

How am I supposed to parse this properly? Sorry if this is very obvious...

[robiningelbrecht]

No one? 😭

[ilvalerione]

You could try to provide the model with stronger instructions using the SystemPrompt: https://docs.neuron-ai.dev/agent#system-instructions

Tell the model the format you want, and even the format you don't want.

[ilvalerione]

@robiningelbrecht I found a bug in the SystemPrompt class that could have an impact on the issue you are facing. You could try with 1.12.11. Let me know it it solves the problem.

[robiningelbrecht]

@ilvalerione ok, I'll keep you up to date, thanks!

[robiningelbrecht]

@ilvalerione Finally got time to get back to this. Seems no matter what I do, the LLM keeps me returning this weird response. My system message:

      return (string) new SystemPrompt(
            background: ['You are an AI Agent specialized analyzing workout results and providing workout tips.'],
            steps: [
                "Answer the user's question.",
            ],
            output: [
                'Write a summary in a paragraph without using lists. Use just fluent text. Do not add any markdown',
            ]
        );

The response:

Any pointers? :p

If it helps, I'm logging everything in inspector.dev. I could share with you the details of an api call...

[ilvalerione]

Ok, you should drop in a live chat on the dashboard so I can check your account data. The answer about what's happening should be there.