Built a unified workspace to debug AI workflows - Looking for feedback
Built a unified workspace for debugging multi-step AI workflows — looking for feedback
I've been building a workspace for investigating AI workflow executions. After spending time with existing observability tools, I kept finding myself jumping between traces, prompts, logs, and metrics. I wanted to see what it would feel like to have the investigation happen in one place and make it easier to know where to start.
The current build has the flow:
Projects -> Sessions -> Runs -> Events
Events can include tool calls, LLM calls, prompts, responses, and other execution details. A run can also exist without a session when there isn't a broader interaction to group it under.
There are filters for things like tool loops and context inflation, along with basic filters for time range and client, to help narrow down where to start. It also captures the business events that happened during the workflow.
For those building AI products or agentic pipelines:
How are you investigating these today?
Even a quick reaction is useful — does this look like something you'd use, or would you approach the investigation differently?
I've been building a workspace for investigating AI workflow executions. After spending time with existing observability tools, I kept finding myself jumping between traces, prompts, logs, and metrics. I wanted to see what it would feel like to have the investigation happen in one place and make it easier to know where to start.
The current build has the flow:
Projects -> Sessions -> Runs -> Events
Events can include tool calls, LLM calls, prompts, responses, and other execution details. A run can also exist without a session when there isn't a broader interaction to group it under.
There are filters for things like tool loops and context inflation, along with basic filters for time range and client, to help narrow down where to start. It also captures the business events that happened during the workflow.
For those building AI products or agentic pipelines:
How are you investigating these today?
Even a quick reaction is useful — does this look like something you'd use, or would you approach the investigation differently?
