Release notes · November 1, 2026 · 5 min read

Nightly 2026-11-01: setup in one page, a grouped footer, faster computer use

v0.0.2-nightly.20261101.18 signs in to ChatGPT from the setup page, lists every model by engine, groups the footer, batches computer-use actions and fixes attach, dictation and sub-agent model names.

Generated editorial cover. Product screenshots inside the post are unedited.

The nightly v0.0.2-nightly.20261101.18 was built on 2026-11-01. It changes the first five minutes with MendCode more than any release so far: setup, the model list and the footer, plus a round of computer-use speed work and fixes that people hit every day.

Install the nightly

A nightly is published each day the main branch changed. The channel you pick is remembered, so later upgrades stay on it.

macOS and Linux

curl -fsSL https://github.com/MendCode/MendCode/releases/latest/download/install.sh | sh -s -- --channel nightly

# already installed
mendcode upgrade --channel nightly

Setup and models

What the setup page does now

ChatGPT sign-in on the page

Setup signs in to ChatGPT itself, the same as /connect chatgpt, instead of sending you to another CLI first.

Every model, grouped

Space on the Model row lists every model, grouped by the engine that runs it.

The runner is named

The Model row says which engine runs the model, for example “Opus · Claude Code”.

The whole terminal

Setup uses the full terminal instead of a small box in the middle.

Model            a reasoning · ↑↓ row · ←→ model · / search · ctrl+a providers · e effort · s fast · esc closenext: GPT-6-Sol · high                                                                                        ┌─ Roles ────────────────────────────────────────────────────────────────────────────────────────────────────┐│                                                                                                            ││ ❯ this session        ‹ GPT-6-Sol › high  gpt-6-sol                                                        ││   main                ‹ GPT-6-Sol › medium  gpt-6-sol                                                      ││   plan                ‹ Sonnet 5 › default  claude-sonnet-5                                                ││   subagent            ‹ same as main ›                                                                     ││   compaction          ‹ same as main ›                                                                     ││   dream               ‹ same as main ›                                                                     ││   aside               ‹ small, else this session's model ›                                                 ││   small               ‹ same as main ›                                                                     ││   permissionReviewer  ‹ same as main ›                                                                     ││   memoryExtractor     ‹ small, else this session's model ›                                                 ││   memoryAssistant     ‹ memoryExtractor, small, else main ›                                                ││   memoryConsolidator  ‹ unset: deterministic Dream consolidation ›                                         ││   title               ‹ small, else this session's model ›                                                 ││   summary             ‹ small, else this session's model ›                                                 ││                                                                                                            │└────────────────────────────────────────────────────────────────────────────────────────────────────────────┘                                                                                                              ┌─ Models ───────────────────────────────────────────────────────────────────────────────────────────────────┐│                                                                                                            ││ ChatGPT subscription  4                                                                                    ││     GPT-6-Luna   gpt-6-luna       low medium high · 272k                                                   ││     GPT-6-Sol    gpt-6-sol        low medium high xhigh · 272k                                             ││     Sonnet 5     claude-sonnet-5  standard · 200k                                                          ││     Qwen 3 480B  qwen3-480b       standard extended · 256k                                                 ││                                                                                                            │└────────────────────────────────────────────────────────────────────────────────────────────────────────────┘

The model picker, drawn from MendCode’s own render code.

Sessions can also read each other now: sessions read=<id> returns another session’s messages in pages of up to 40,000 bytes. The start screen draws the infinity logo in half blocks.

The footer reads in groups, and the reasoning effort is shown by default:

footer

Build · GPT-6-Luna · high · fast · native │ ↻ 1 loop │ ~12 turns to compact

Mode, model, effort and engine first; then running loops; then how far the session is from compaction.

Computer use

What changed

tools.batch runs up to 16 actions in one call.
The cursor comes from agent-desktop (Apache-2.0). A move takes 90–320 ms instead of 280–750 ms.
Observing a window is 6–7× faster: Ghostty’s 239 nodes went from about 500 ms to about 65 ms.
A failed action says whether it was not delivered or whether delivery is unknown, so the agent does not repeat a click that may have landed.

Fixes

Fixed in this nightly

mendcode attach <id> works again.
Dictation listens in two languages and shows words about one second after they are said.
Sub-agents take models by the names mendcode models prints, such as chatgpt/gpt-6.1-terra, claude/claude-opus or codex/codex:gpt-6-sol.
The thread bridge names MendCode’s models.
There is no hidden summary call per turn any more; it could cost up to about 6k tokens. Titles and summaries now go through the budget.
Changing the default model in the middle of a session no longer breaks it with 400 model_not_found.