Workspace chat — describe the goal, expand a tool step
Workspace
Configure in conversation
BSW parameters and ASW modeling—describe the outcome, skip the menu tree.
1Goal-first — Skip deep menu trees—describe the outcome you want.
2Add or drop modules — Say "add Can" or "remove Det"—the agent enables or removes them in the project, not in the configurator UI.
3ASW as well — Create and edit data types, interfaces, ports, SWCs, and internal behavior in natural language; wire ports and map signals.
4AUTOSAR-aware — Understands BSW modules and ASW components—constraints, conventions, and your stated goals; never alters the project arbitrarily.
Component diagram
Through your existing toolchain
Edits, verification, and code generation go through the configurator you already have. Oversized projects load per module.
1Through your toolchain — Edits, verification, and code generation go through the configurator you already have. It does not edit ARXML as text.
2Iterates on errors — When verification fails, the agent reads toolchain errors and keeps adjusting instead of dumping logs on you.
310M+ ARXML — Oversized ECUC files load per module; chunked bulk edits and in-project scans—what generic coding agents cannot do.
Component diagram — search, select, and zoom connections
Configuration changes — inspect by module and path, then undo
Configuration changes
Logged, replayable, undoable
Edits that never hit a tool do not count. Every legitimate write is logged; pick one and roll it back.
1No phantom edits — Changes that never hit a tool do not count. Every legitimate tool write is logged—reviewable by time or URL. View sample edit table →
2Replayable edits — Every tool-applied change is logged by time and path—open the list and see exactly what moved.
3Undo a change — Pick one entry and roll it back; the project returns to the prior state without hand-editing.
Plan mode
Align on requirements, then change config
For multi-step work, inspect the project first and write a checkable plan. Agree on goals and steps before any write.
1Align — Match your goals to the live project—what to change, and how far.
2Write the plan — Goals, structure, and steps in one document you can review on the side.
3Then execute — Start only after you confirm. Todos track progress, including verify and generate.
Turn requirements into a plan; todos follow the steps
Same task, same prompt: finish the configuration in one conversation, then upload the project and score it row by row against that task's published rubric. Tasks and criteria are public, so you can compare assistants on whether they actually get the configuration through.
1
One prompt
Finish in that conversation; no follow-up rounds.
2
Upload config
Each row is pass or fail. Missing paths do not pass.
3
Follow the task
How to judge is on the task; not every task uses the same check.
