editor-info

Install from npm:

pi install npm:pi-editor-info

Reload Pi after installation:

/reload

Custom pi editor shell:

──────────────────── gpt-5.4 fast • xhigh • ▰▰▱▱▱▱▱▱▱▱ 12.3%/128K • 2.0K tok/s
❯ Type a prompt here
─────────────────────────────────────────────────────────────────────────────

• Top and bottom borders use the fixed neutral-gray muted theme color. They do not change with thinking level.
• Pi's built-in footer is replaced with an empty component because its model, context, token, cost, and directory information is redundant with the editor border. It is restored automatically when the extension shuts down.
• ❯ is the input-start marker.
• Submitted user messages keep pi's existing userMessageBg box and add the same accent-colored ❯ prompt inside it. Wrapped and multiline content uses a two-column continuation indent.
• Top-right information order is model, thinking level, context usage, and current-model token throughput.
• Context usage uses a modern segmented ▰/▱ progress bar with no ctx prefix. It displays only percentage and total context-window size, for example ▰▰▰▰▰▰▱▱▱▱ 63.7%/128K.
• GPT model labels append fast only while fast mode is active, read from @pi-plugins/fast-mode's shared status registry. No fast label is shown when the mode is off.
• The redundant [fast mode] row above the editor is filtered out while the registry data remains available for the top-border fast label. Other extensions using the shared pi-plugins:statusline row are preserved.
• Throughput is time-weighted across the last eight completed assistant messages. While streaming, the live estimate uses incremental token deltas and a time-aware EWMA (rather than cumulative tokens divided by total response time), so it is less sensitive to time-to-first-token and provider update bursts. It resets on model change.
• All normal CustomEditor behavior remains available: history, multiline input, autocomplete, paste markers, IME cursor placement, and app keybindings.