Evidence record / pi-coding-agent
A dated record of one change, kept so the writing that cites it can be checked. Compiled from the sources listed below by the research run, not written for reading. The judgment lives in the issues that cite it, below.
2026-08-10-pi-coding-agent-fullscreen-tui-mode-mermaid-and-latex-rendering-baseten-and-qwen-token
Fullscreen TUI mode, Mermaid and LaTeX rendering, Baseten and Qwen Token Plan providers.
v0.84.0 adds a fullscreen TUI mode switchable at runtime, with a sticky editor and footer, an independently scrollable transcript, and draggable scrollbars; Mermaid diagram and terminal-Unicode math rendering in interactive transcripts; arbitrary OpenAI-compatible samplingParams plus opt-in vLLM thinking_token_budget; and a built-in Baseten provider. v0.84.1 adds a Qwen Token Plan Individual provider (PR #7659, @arasovic) and multi-click word/paragraph selection. v0.84.2 adds fullscreen transcript search on Ctrl+Shift+F, a configurable fullscreen exit output (print the transcript, or only a resume hint), and --use-theme <name[/name]> for a per-run theme that does not touch saved settings (PR #7722, @rwachtler). Note the qualifier in docs/usage.md at v0.84.2: the flag is documented as --tui-mode <mode> with regular (default) or experimental fullscreen, and iTerm2 renders inline images only as text placeholders in fullscreen because its inline-image protocol cannot crop placements during application-owned scrolling.
Channel: tagged-release. Ancestry: Documented in the v0.84.0 and v0.84.1 release bodies. v0.84.0 tag sha a5f43bf8aff3c55752432655f7334e3dafd1e256, v0.84.1 tag sha 53fa77ccd8a279eb87e92294ef3687b03ff80112, both prerelease=false and draft=false per gh api repos/earendil-works/pi/releases. Supporting commits appear in gh api repos/earendil-works/pi/compare/v0.83.0...v0.84.2: 05e89b41 "feat(tui): render LaTeX math in Markdown" (2026-08-05), 5446cd75 "feat(coding-agent): rename UI mode to TUI mode" (2026-08-05), c03d78bd "feat(ai): add Qwen Token Plan Individual provider (#7659)" (2026-08-06).
Operator consequence: Try fullscreen mode, but keep regular as the default for now. The three releases spend most of their surface area on the terminal, and the payoff is real for long sessions -- a transcript you can scroll and search without losing the editor. The caveat is that the release notes call it a headline feature while the docs still call it experimental, with named terminal-specific degradations. That gap between the release note and the doc is the thing to notice. samplingParams and thinking_token_budget matter more if you self-host: they let a vLLM or OpenAI-compatible endpoint be tuned from Pi's config instead of forked code.
Receipt
Finding metadata
Run: 2026-08-10-weekly-digest-2026-08-03_2026-08-10-frontier-v0
Finding ID: 2026-08-10-pi-coding-agent-fullscreen-tui-mode-mermaid-and-latex-rendering-baseten-and-qwen-token
Source links
Primary links, including exact changelog lines when available.
Versioned source: run artifact