docs/compatibility.md records AWS Databricks as not tested, and the README's verification
table says the same. Every path has only ever run against an Azure workspace.
This is listed as untested rather than assumed to work because identical API paths across clouds
are not evidence — that sentence is already in compatibility.md:6 and this issue exists so it can
eventually stop being true.
What closing this looks like
Anyone with an AWS Databricks workspace and a Genie Agent can do it:
export DATABRICKS_HOST="https://<your-workspace>.cloud.databricks.com"
export DATABRICKS_TOKEN="<token>"
export LAKESPEAK_LIVE_AGENT="<your agent name>"
dotnet test -c Release --filter "Category=Live"
That is 9 tests covering agent listing, a question and its SQL, decimal fidelity, follow-up
conversations, unknown-agent handling, feedback, re-execution and cancellation.
Then add a row to the Clouds table in docs/compatibility.md and the README's verification
table with the date and what was run. Please record what failed too, if anything — an entry with
no evidence behind it is worse than a missing one.
Worth knowing before you start
The live suite runs real questions against a real SQL warehouse, so it costs warehouse time. It
does not write anything.
docs/compatibility.mdrecords AWS Databricks as not tested, and the README's verificationtable says the same. Every path has only ever run against an Azure workspace.
This is listed as untested rather than assumed to work because identical API paths across clouds
are not evidence — that sentence is already in
compatibility.md:6and this issue exists so it caneventually stop being true.
What closing this looks like
Anyone with an AWS Databricks workspace and a Genie Agent can do it:
That is 9 tests covering agent listing, a question and its SQL, decimal fidelity, follow-up
conversations, unknown-agent handling, feedback, re-execution and cancellation.
Then add a row to the Clouds table in
docs/compatibility.mdand the README's verificationtable with the date and what was run. Please record what failed too, if anything — an entry with
no evidence behind it is worse than a missing one.
Worth knowing before you start
The live suite runs real questions against a real SQL warehouse, so it costs warehouse time. It
does not write anything.