Tale v0.5.105

Highlights

• Agents on Pi keep working through long conversations. Tale did not tell Pi how much context a model holds, so Pi took 128,000 tokens for every model it reached through Tale. Once a conversation came near that, Pi asked the model for a one-token answer and the run ended without a report ("the agent ended its turn without a final report"); a retry continued the same conversation and ended the same way. A long task stopped there, and so did a standing role whose runs continue one conversation. Pi now takes each model's context window from the model catalog, narrowed by the organization's context limit, answers in full and compacts the conversation before it outgrows the window. A model whose window Tale cannot resolve keeps Pi's own sizing.

Upgrade notes

• Platform: run the 0.5.105 images. No configuration change and no migration. An agent on Pi sizes its conversation to its model's window from its next run, so a task whose run ended this way can be started again and continues its conversation within that window.

API contract changes

None in this range. The contract stays at 3.53.0: 163 operations.

What's Changed

• fix(platform): give agents on Pi the model's context window by @yannickmonney in #5416

Full Changelog: v0.5.104...v0.5.105