Bagel 2.0: plain-English rosbag queries with smart auto-reduction

Bagel 2.0 from Extelligence-ai lets robotics engineers talk to their bag files instead of writing one-off scripts. Point Claude or any MCP client—including fully local models via Ollama—at a bag and ask natural-language questions like “what was the peak z-deceleration?” or “find every hard brake and cut ±10s snippets.” The tool doesn't rely on LLM guesses: it writes topic messages to Arrow and runs DuckDB SQL over them, displaying the exact query so users can audit the answer. In testing, the team tried to bait it into confirming a pothole that wasn't in the data; it checked the z-axis, told them the data was flat, and explained why—a tool that says "no" when the answer is no was a hard requirement.

The headline feature is the data reduction pipeline. A single sentence like “keep 10 seconds around every deceleration below −10 m/s²” becomes a detector that previews exactly what it would keep before writing a byte. It can run on one bag, across a fleet, or even on the robot itself, so only event windows are ever written—with message-level verification ensuring kept windows are byte-identical to the source and nothing outside them survives. The release also expands format support: ROS 1/2 bags, MCAP (any profile), ROS text logs, PX4, ArduPilot, Betaflight, CAN/MF4, live MQTT, and Copper logs, with exports to Rerun, Lichtblick, PlotJuggler, and LeRobot datasets. Everything runs in Docker locally, with no cloud and no telemetry, under an Apache-2.0 license.