Novelist 0.2.7 — a more controllable AI Agent
AI Talk and AI Agent now share an explicit context composer: attach selections, the current file, outlines, or project files with @ mentions, then review AI Agent edits in Apply Changes cards before accepting them file by file.
0.2.7 focuses on the parts of AI writing that most need control: what the assistant reads, which context is sent, and what files it wants to change.
Explicit context
AI Talk and AI Agent now share a compact composer. Use @ to attach exactly the context you want:
@selection: the current editor selection@current: the active file@outline: the document structure@file/@folder: project files and folders
Attached context appears as removable chips before you send. Highlighted text is no longer silently injected into every prompt; it becomes a visible suggestion you can accept or ignore.
Apply Changes
When AI Agent returns a novelist-change-set block, Novelist renders it as a review card. You can accept, reject, copy, or regenerate per file, and stale originals are called out before you apply a change.
The Agent now behaves more like a writing partner proposing a patch than a black box rewriting text out of sight.
Cleaner conversation payloads
Attached files and snippets are summarized once in the transcript and then excluded from duplicated prior-turn payloads. Long AI sessions stay cleaner and less likely to drift because the same context was repeated too many times.
Release reliability
The release pipeline also picked up two important fixes:
- The release job no longer gets skipped by GitHub Actions’ implicit success gate when one platform matrix leg fails.
- Intel macOS DMGs are now cross-compiled from an Apple Silicon runner, avoiding long queues on retiring GitHub-hosted Intel runners.
Those changes are mostly invisible in daily writing, but they make future releases sturdier.
Also
- Full details: v0.2.7 CHANGELOG
- GitHub Release: Novelist v0.2.7