AI Agent
An autonomous ReAct agent that picks its own tools to reach a goal, showing every step it takes along the way. Its operational tools run against live MongoDB, PostgreSQL and Redis instances; the public demo below runs the JSON repair tool, so you can watch the reasoning loop work without it touching stored data.
How it thinks
Tools it can call
Repairs malformed JSON and explains what was wrong with it.
Checks connectivity and health across every connected database at once.
Runs find, count, aggregate and listCollections against MongoDB.
Runs read-only SQL — only SELECT and EXPLAIN statements are permitted.
Lists keys by glob pattern, reads values, and inspects type and TTL.
Reports Redis server stats, memory usage and connection details.
Summarises AI token usage and cost by user, endpoint or date range.
Try asking
How this demo is sandboxed. Every tool is read-only to begin with — nothing can write or delete, and SQL is restricted to SELECT and EXPLAIN. Beyond that, public access is allowlisted rather than filtered: the only tool reachable without an admin session is JSON repair, which touches no datastore at all. Access is decided server-side on every tool call and defaults to denied, so the boundary holds even if the model is persuaded to attempt something else — and blocked tools are never shown to it in the first place. Each reply reports the model used along with its token count and cost.