Last week I launched AgentGrove, an open-source local workspace that runs AI coding agents (Claude Code, opencode, Kimi) in isolated git worktrees. Today's release, v0.1.42, makes its database tooling much more useful: a DBeaver-style database tree.
What's new
The left rail's Database view now renders the full connection → databases → tables tree:
- Every database on the server, expandable and lazy-loaded — not just the one in your connection string
- One click switches the active database; the SQL editor and table browser immediately target it (the pane header shows which DB you're on)
- A single filter box matches database names and table names
- Saved connections with an inline Test button (server version on success, the real error on failure)
- SQL editor with schema autocomplete — tables and columns of the active database, prefetched in the background, on top of CodeMirror with Ctrl/Cmd+Enter to run
- Table browser with column filters (=, !=, <, >, LIKE, ILIKE) and pagination
Under the hood the backend is a small Rust/tokio-postgres layer: SELECTs are wrapped with to_jsonb(...) so any row shape renders as JSON, and cross-database browsing is just the connection URL's database segment swapped per request — no extra connection state to manage.
Why it matters here
Agents write a lot of SQL-shaped code, and debugging their output means looking at actual data. Having the database tree a click away from the agent chat — instead of alt-tabbing to DBeaver — keeps the review loop in one window: agent writes a migration, you inspect the affected tables right next to the diff.
Try it
git clone https://github.com/arnabk/agentgrove.git
cd agentgrove && just dev
Point it at any Postgres (the first-run seed targets your local one; AG_DATABASE_URL overrides). Demo videos, including the DB editor, are on the README. What should the database view do next — edit rows inline? Run agent-proposed migrations with a one-click apply?
Top comments (3)
Update: AgentGrove is now at v0.1.53. Since this post: the database view added Postgres support with a full connection to databases to tables tree and SQL schema autocomplete; a new Material Dark theme plus user-editable custom themes; chat gained fork-from-point, per-message retry/delete-to-point, and model-switch context handoff; git worktrees got a zoomable "Galaxy Map", auto-generated names, and restore-with-chat-recovery; worktrees are now first-class in the left rail with "open folder in OS file manager"; terminal rendering fixes and bounded scrollback; plus a client memory-growth monitor. Repo: github.com/arnabk/agentgrove (v0.1.53).
A database tree inside an agent workspace is more than convenience. It gives the human and the agent the same map of what exists, which reduces the chance that the agent invents schema assumptions from memory.
Author's note: the title of this post mentions another database tool by name for comparison; to be precise, the database tree described here is AgentGrove's own built-in feature (connection → databases → tables, with SQL autocomplete) and isn't affiliated with or endorsed by any other product.