v0.1.0

The first release of Escrowe: an LLM agent writes SQL against your database without ever seeing the data. It sees the connected account's schema only, its SQL runs as that account, and the rows come back to you, never to the agent.

Highlights

• Databases: MySQL, PostgreSQL, SQL Server, SQLite, DuckLake and Iceberg REST catalogs. Several saved databases, one active at a time.
• LLMs: Anthropic or OpenAI with an API key, or the Claude Code and Codex CLIs reusing your existing browser login.
• Blind by design: the agent gets the schema and each result's shape (row counts, all-NULL columns), never rows, unless a person explicitly shares a result with \feed. A guard keeps the agent to a single read-only query.
• Audit and transcript: every attempt is logged, and every prompt sent to the LLM can be inspected with escrowe llm-log.
• CLI and browser UI: an interactive prompt (escrowe) and a notebook interface (escrowe -ui) with SQL, question and text cells, charts, and a present mode.
• In the notebook: the agent's SQL is highlighted and editable, and an edited query becomes the context for the next question. A Clear context button starts fresh. Charts let you choose which columns the tooltip shows.

Install

pipx install "git+https://github.com/bureaugewas/Escrowe.git@v0.1.0"

Python 3.10 or newer.