Hi HN, I’m the founder of an open-source infra access gateway. I’ve always been frustrated by how our users were missing the cool data generated by our tool.
Think a list of all actions devs perform on infrastructure: SQL, shell, CLIs, etc. I dreamed of the day our tool would be able to tell users: “hey, you run this query every week, consider adding it to an internal tool or scheduler”. Or: “you read 1000 customer e-mails last week, likely a mistake, please remove this column from your queries.”
We tried building it a few times. First indexing session contents on Elasticsearch. Then trying to query directly from Postgres blob data types in-line parsing to json. All failed. Data is too big and making this work would be it’s own product (SIEM?).
Thanks to LLMs we were able to launch something much simpler that does a better job then I ever imagined: an MCP server where users can query their session history from an agent.
Agents do a good job of pulling only the right pieces of sessions and analysing them in small chunks to surface cool ideas. And users aren’t locked to whatever rules we add to our interface, they can freely ask the agent about the stuff they care about.
Think a list of all actions devs perform on infrastructure: SQL, shell, CLIs, etc. I dreamed of the day our tool would be able to tell users: “hey, you run this query every week, consider adding it to an internal tool or scheduler”. Or: “you read 1000 customer e-mails last week, likely a mistake, please remove this column from your queries.”
We tried building it a few times. First indexing session contents on Elasticsearch. Then trying to query directly from Postgres blob data types in-line parsing to json. All failed. Data is too big and making this work would be it’s own product (SIEM?).
Thanks to LLMs we were able to launch something much simpler that does a better job then I ever imagined: an MCP server where users can query their session history from an agent.
Agents do a good job of pulling only the right pieces of sessions and analysing them in small chunks to surface cool ideas. And users aren’t locked to whatever rules we add to our interface, they can freely ask the agent about the stuff they care about.
Project is here if you want to give it a shot: https://github.com/hoophq/hoop
Does this make sense to you? Would help a lot even if you told me it doesn’t.