More control over agent runs
Agent runs are the shortest path from issue to diff. Assign a task to the Plinth agent and it works in an isolated environment to write and test the change before opening a pull request.
We are adding more control over how each run behaves, from its environment to the models behind it.
Adaptive routing
Save time and credits by routing simpler tasks to a faster model while keeping the default reasoning model for complex work. If a task needs more than expected, restart the run on the default model and keep the original pull request for reference.
Environment secrets
Runs can now read environment secrets during setup, so they can fetch private dependencies and other configuration safely.
Bring-your-own models
Runs now support self-hosted models. Workspace admins choose which models are available from Agent settings.
Fixes
Fixed an issue where the board scrolled to the top after moving a card
Due dates now respect the workspace time zone in exported CSVs
Resolved a crash when pasting very large tables into a doc
Improvements
Faster initial load for workspaces with more than 50 teams
Cycle reports now include carried-over scope
The command menu remembers your last three actions