Let your agent write a table, or upload a file. rho keeps every version, so any agent can pick the table up next week, undo a bad change, and share it as a link or an API.
https://api.rho.store/mcpKeep
Your agent saves what it finds as a table with typed columns, under a name. Next week, another chat or another agent opens it and adds to it. A big table comes back as a summary and a sample, so the agent isn't swamped.
Refresh the licences in our vector database table.
Claude Code
Updated 2 rows and added 1. Version 2.
write_table("research/vector-dbs", rows, mode="upsert", key="name")Undo
Every write is a new version, and old versions never change. A bad change is one restore away, and the restore is a version too. A delete goes to the trash.
That refresh was wrong. Put it back.
Claude Code
Restored version 1 as version 3. Nothing was lost.
restore_version("research/vector-dbs", version=1)Share
Make a table public and anyone with the link can open it, without signing in, and code can read it as an API. A figure can come with a receipt that shows the SQL behind it.
Send this to my colleague.
ChatGPT
Here's a link anyone can open.
share_table("research/vector-dbs")Drop a CSV, Excel, JSON or Parquet file into rho and its table has a URL: rows as typed JSON, with filters, sorting and paging, and no server to run. A new upload is a new version, so the API doesn't change under the code that calls it. From Python, one call saves a DataFrame with its types.
curl https://api.rho.store/v/<key>?country=SE&sort=-pricerho.put(df, "research/sales")A sheet is made for people editing cells. rho is made for agents writing tables.
The agent writes rows with typed columns. There's no sheet layout for it to keep straight.
It reads a summary and a sample, then asks SQL for the rest, rather than reading every cell.
Each write is a version, and the agent restores one in a single call.
Share the table with a person, or let code read it, from the same place.
If people do most of the editing by hand, a spreadsheet is the better tool.
rho is an MCP server. Add its URL, sign in once, and your agent has its tools. There's no code to write.
Writing your own? There's a Python SDK, and every table has its ownAPI.
| Research your agent should keep and grow | rho |
| Results to pass between agents and people | rho |
| A file you want as an API | rho |
| A sheet people mostly edit by hand | a spreadsheet, such as Google Sheets |
| A one-off answer you won't come back to | a CSV in the chat |
| The database behind your app | a Postgres, such as Neon or Supabase |
| An agent's memory of text and facts | a memory store, such as Mem0 or Zep |