Download Latest Version v0.2.13 source code.zip (92.1 MB) Google Add to Preferred Sources
Home / v0.2.13
Name Modified Size InfoDownloads / Week
Parent folder
ante-v0.2.13-darwin-arm64.tar.gz 2026-10-19 14.0 MB
ante-v0.2.13-darwin-x86_64.tar.gz 2026-10-19 15.5 MB
ante-v0.2.13-linux-arm64-musl.tar.gz 2026-10-19 14.7 MB
ante-v0.2.13-linux-x86_64-musl.tar.gz 2026-10-19 16.1 MB
README.md 2026-10-19 2.3 kB
v0.2.13 source code.tar.gz 2026-10-19 91.9 MB
v0.2.13 source code.zip 2026-10-19 92.1 MB
Totals: 7 Items   244.2 MB 0

v0.2.13 - 2026-10-18

Added

  • Jump-to-bottom pill in fullscreen: while the conversation is scrolled up, its last row shows N new messages (ctrl+l) ↓, or Jump to bottom (ctrl+l) ↓ when nothing new has arrived; Ctrl+L or a click on that row jumps to the latest output, and sending a prompt or a !command follows new output again
  • Subagents record their own event log under the parent session, at sessions/<parent>/subagents/<child>/events.jsonl; a subagent is not a resumable session, so /resume does not list it, and deleting the parent removes its logs. The parent's Agent tool result names the child's session id
  • A warning when session history cannot be saved: a failed event-log or snapshot write shows one notice per session with the path and disk-space/permission guidance, and the conversation keeps running
  • Gemini 4 Argon on Antix; DeepSeek V4.1 Flash 1017, Reflection Beam 1.0 and Muse Spark Open in the OpenRouter catalog

Changed

  • Compaction's fold call gets room for reasoning: its output cap is twice the summary's budget, bounded by the model's declared max_tokens and the context window, so a model that thinks before summarizing no longer has its summary cut off at the cap
  • The fold runs at reasoning effort low at most (min stays min; a model without an effort setting stays without one), since a summary does not need the session's effort
  • Dependency updates

Fixed

  • A Responses API response.incomplete with reason length is treated as a max-tokens stop: a cut-off compaction summary reports summary was cut off at N tokens instead of an opaque failed to fold dialog, and a normal turn continues as it does after max_output_tokens
  • Interrupting a turn before the model produces any output attaches the Interrupted row to the prompt you just sent instead of the previous answer, and Ctrl+C / Esc between send and the turn's start interrupts the coming turn instead of acting as if the session were idle
  • In fullscreen, a bare click on the conversation (press and release without dragging) no longer stops the view from following new output; a drag selection still holds the view in place

Crates

  • Public crates published as 0.2.13: ante-protocol-shape, ante-llm, ante-exec, ante-sdk
Source: README.md, updated 2026-10-19