Exploration & Conversation

Summary

The exploration interface is the primary way users interact with their data. It is an agent-driven conversational UI where the user asks questions, and agents search, analyse, and return findings from the user's ingested documents and data.

Current State (from Condelo RAG)

The existing chat interface in condelo-rag provides:

Threaded conversations with streaming responses
Tool calling (document search, text-to-SQL, web search, sub-agents)
Tool status display showing what the agent is doing
Nested sub-agent analysis with collapsible output

What Carries Over

The core exploration mechanics come from the platform API — Artefaq calls the platform for agent conversations, tool execution, and streaming responses.

Next Steps / Suggested Actions

After an agent returns a response, the user is presented with:

Next steps — suggested follow-up questions or actions based on the response
Suggested background tasks — longer-running analyses the agent can kick off

This pattern exists in the current app but needs refinement:

How are next steps generated? (LLM-suggested vs. template-based vs. hybrid)
How many should be shown? (3–5 feels right, but needs testing)
Should they be categorised? (e.g. "Dig deeper", "Related topic", "Take action")
How do background tasks report back? (Notifications? Inline update?)

Post-Response Actions

At the end of each agent response, the user sees action buttons:

Bookmark — save this point in the exploration (see 07-bookmarks.mdx)
Save to Workspace — extract content as a snippet (see 03-workspace-snippets.mdx)

These need to be prominent but not intrusive — they should feel like a natural part of the conversation flow.

Open Questions