Dashboard sub-feature: CAgent Observability

[marcellodesales]

What you'd like to see

• Having multiple agents involves getting into the Observability layers to find out how agents are talking to
  • What is the consequence of the prompt set in my agent?
  • Are the assigned MCP Toolsset correctly being used?
  • What improvements can I provide to the Prompt used for a better memory metadata utilization? What if it's correlating incorrect data?
  • Is the MCP Toolsset used requiring Authentication for upstream API servers? How to set them up properly?
  • If the context window is a bottleneck for my Agent, how can users identify the agent's setting to improve its performance?

Why you'd like to see it

• A basic way to debug Cagent, showing the Agent Prompt, Toolsset used, and Memory and Brain
  • Prompt: it can show the current prompt, estimates on its setting such as context windown, etc. Screen with "Test" env that allows developers to quickly verify specific prompt changes in the behavior
  • Toolsset: a view on the set of the MCP Servers used, through the Docker MCP Gateway (feature-rich) or others (feature-simple) where developers can verify, whitelist, disallow offered tools. In addition, developers can see any egress profile of the MCP Server, making sure it's not an unsafe server. That is, how to safely profile MCP servers in a way that it doesn't deviate from their normal behavior.
  • Memory: Making sure that any changes to the prompt and lifecycle of an agent is versioned. That is, what if the memory model used is Unstructured and tomorror we want to replace it with a Knowledge Graph? What if I want to experiment the model of Whiteboard memory, where agents reuse a single memory model in order to reason in a "whiteboard" style?
  • Brain: the choice of the models and how they impact the agent's performance as described above.

Workarounds?

• Still going through cagent's usage and operational model, I see myself operating one or two agents. However, the the number increases, I see the need of a more holysitc view of the Agent's world.

Additional context

This is the expansion of the ticket #417, where the Dashboard concept is the visual while the Observability is the way to get into cagents internal model.

[krissetto]

I'm a bit confused as to what's desired here? Better observability in general? (which i agree with)

Could you provide a more practical example of the experience you're imagining? It'd help us in triaging this issue. Thanks!