From 5902078a5917ec58e21d86b90c46e9b7ab9dfa4e Mon Sep 17 00:00:00 2001
From: Jacob Fu <141651335+FuJacob@users.noreply.github.com>
Date: Mon, 1 Jun 2026 22:59:29 -0700
Subject: [PATCH] Match Extended Context cap to what the engine actually uses

The UI advertised a 4000-character Extended Context limit, but the Open Source base-model path rendered it as a 'notes' section hard-capped at 600 chars (including a ~32-char label), so only the first ~570 chars were ever used on-device and the rest was silently dropped.

Lower the user-facing/storage cap to 1200 and raise the notes section cap to 1300 so the full entry reaches the local model. The total base-prompt budget stays 2400 chars, so prompt size and latency are unchanged; this only stops wasting the budget. Both caps now document the invariant (notes maxChars >= UI cap + label) so they cannot silently diverge again.

Also shrinks the Foundation Models nano context-window overflow described in #486.

Refs #486
---
 Cotabby/Models/SuggestionSettingsModel.swift      | 15 +++++++++------
 .../Support/BaseCompletionPromptRenderer.swift    |  7 ++++++-
 2 files changed, 15 insertions(+), 7 deletions(-)

diff --git a/Cotabby/Models/SuggestionSettingsModel.swift b/Cotabby/Models/SuggestionSettingsModel.swift
index 20c1009..46e3f97 100644
--- a/Cotabby/Models/SuggestionSettingsModel.swift
+++ b/Cotabby/Models/SuggestionSettingsModel.swift
@@ -138,12 +138,15 @@ final class SuggestionSettingsModel: ObservableObject {
     static let defaultGhostTextOpacity: Double = 1.0
     static let ghostTextOpacityStep: Double = 0.1
 
-    /// Hard upper bound on the persisted Extended Context blob, in characters. Sized so the user
-    /// can paste a meaningful glossary or style guide without crowding the model's shared context:
-    /// roughly ~1000 tokens of English, which still leaves headroom for instructions, prefix text,
-    /// clipboard, and visual context inside Apple's 4096-token window. Larger pastes are truncated
-    /// at write time so the cost is bounded on every subsequent request.
-    static let maximumExtendedContextCharacters: Int = 4_000
+    /// Hard upper bound on the persisted Extended Context blob, in characters. Sized to match what the
+    /// engines actually consume rather than what they can store: the OSS base path renders this as a
+    /// budgeted "notes" section (`BaseCompletionPromptRenderer`, `maxChars` 1300) inside a 2400-char
+    /// prompt, so a larger cap would just be clipped on-device instead of used. ~1200 chars (~300
+    /// tokens) is a meaningful glossary or style guide that still leaves room for the prefix and other
+    /// context, and stays well inside Apple's 4096-token window on the Foundation Models path. Keep this
+    /// at or below the notes section's `maxChars` minus its label so the full blob survives on the OSS
+    /// path. Larger pastes are truncated at write time so the cost is bounded on every request.
+    static let maximumExtendedContextCharacters: Int = 1_200
 
     init(
         configuration: SuggestionConfiguration,
diff --git a/Cotabby/Support/BaseCompletionPromptRenderer.swift b/Cotabby/Support/BaseCompletionPromptRenderer.swift
index f92ac62..dd1f5dc 100644
--- a/Cotabby/Support/BaseCompletionPromptRenderer.swift
+++ b/Cotabby/Support/BaseCompletionPromptRenderer.swift
@@ -43,7 +43,12 @@ enum BaseCompletionPromptRenderer {
             sections.append(Self.contextSection("language", language, priority: 50, maxChars: 300))
         }
         if let notes = Self.nonEmpty(extendedContext) {
-            sections.append(Self.contextSection("notes", "Notes the writer keeps in mind: \(notes)", priority: 40, maxChars: 600))
+            // `maxChars` must stay at or above `SuggestionSettingsModel.maximumExtendedContextCharacters`
+            // plus this label (~32 chars) so the full user-entered Extended Context survives here instead
+            // of being silently clipped far under the advertised cap. It still competes for the 2400-char
+            // total budget below (priority 40), so an unusually long prefix can trim it, but in normal use
+            // the whole blob lands.
+            sections.append(Self.contextSection("notes", "Notes the writer keeps in mind: \(notes)", priority: 40, maxChars: 1300))
         }
         if let clip = Self.nonEmpty(clipboardContext) {
             sections.append(Self.contextSection("clipboard", "On the clipboard: \(clip)", priority: 35, maxChars: 400))