Hi everyone,
Genie Agent Mode has been incredibly impressive and is one of the best tools weโve come across, especially in how it handles multi-step reasoning and delivers structured end-to-end insights.
Weโre currently building a frontend application and would love to use Genie Agent Mode as the backend engine. However, weโve hit a key limitation:
Agent Mode is not available via API, and its capabilities are restricted to the UI.
What we need
To integrate this into a production application, weโre specifically looking for:
1. Agent Mode API access
- Ability to send prompts programmatically
- Execute full Agent workflows outside the UI
2. Reasoning trace / intermediate steps
Structured breakdown of the agentโs internal process, including:
- Question understanding
- Query generation
- Iterative refinement steps
- Final answer synthesis
- Additionally, lightweight status updates during execution such as:
โUnderstanding questionโฆโ
โRunning queriesโฆโ
โAnalyzing resultsโฆโ
These intermediate updates are extremely valuable for frontend UX and transparency.
3. Streaming support (real-time response)
- Responses delivered in incremental chunks (similar to a typing effect) rather than a single final response
- Ability to stream partial outputs to the frontend
- This allows users to:
- See answers forming in real time
- Stay engaged during longer-running queries
- Combine streamed output with live status updates
Genie Conversation APIs are helpful, but they donโt replicate the multi-step reasoning and iterative behavior of Agent Mode.
This would unlock a huge opportunity to build real-time data applications directly on Databricks.
Would love to know if this is on the roadmap.
Thanks!
