Manage the KV cache disc usage from preferences and stats

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

src/engine.rs

@@ -313,6 +313,9 @@ impl Model {
 pub(crate) struct Generator {
     executor: metal::Executor,
     checkpoint: Option<PathBuf>,
+    /// Token frontier of the last transient store, so continued checkpoints are
+    /// spaced like ds4's `continued_last_store_tokens`.
+    last_store_tokens: u32,
     metrics: Arc<Metrics>,
 }
 
@@ -358,6 +361,7 @@ impl Generator {
         Ok(Self {
             executor,
             checkpoint: None,
+            last_store_tokens: 0,
             metrics,
         })
     }
@@ -414,7 +418,7 @@ impl Generator {
         let history = messages
             .split_last()
             .map_or(messages, |(_, history)| history);
-        let store = KvStore::open(directory)?;
+        let store = KvStore::open(directory, settings.kv_cache.budget_bytes)?;
         let history_key =
             conversation_key(&settings.system_prompt, settings.reasoning_mode, history);
         let history_tag: [u8; 32] = Sha256::digest(&history_key).into();
@@ -423,14 +427,17 @@ impl Generator {
             if let Some(entry) = store.find(&history_key, self.executor.context()) {
                 if self.select_checkpoint(&entry.checkpoint, entry.tag)? {
                     store.touch(&entry)?;
+                    self.last_store_tokens = entry.tokens;
                     previous_checkpoint = Some(entry.checkpoint);
                 } else {
                     store.discard(&entry);
+                    self.last_store_tokens = 0;
                     previous_checkpoint = None;
                 }
             } else {
                 self.executor.reset()?;
                 self.checkpoint = None;
+                self.last_store_tokens = 0;
                 previous_checkpoint = None;
                 self.metrics.kv_lookup(KvLookup::Miss);
             }
@@ -450,6 +457,17 @@ impl Generator {
             let completed_key =
                 conversation_key(&settings.system_prompt, settings.reasoning_mode, &completed);
             let completed_tag: [u8; 32] = Sha256::digest(&completed_key).into();
+            if !settings.kv_cache.stores(
+                self.executor.position(),
+                history.is_empty(),
+                self.last_store_tokens,
+            ) {
+                // A gated store still leaves the finished conversation in the
+                // live KV, so mark it and drop the stale file association.
+                self.executor.note_checkpoint_tag(completed_tag);
+                self.checkpoint = None;
+                return Ok(output);
+            }
             let completed_checkpoint = store.checkpoint_path(&completed_key);
             self.save_checkpoint(&completed_checkpoint, completed_tag)?;
             let retained = store.record(
@@ -468,6 +486,9 @@ impl Generator {
                 .then(|| std::fs::metadata(&completed_checkpoint).ok())
                 .flatten()
                 .map_or(0, |item| item.len());
+            if retained {
+                self.last_store_tokens = self.executor.position();
+            }
             self.checkpoint = retained.then_some(completed_checkpoint);
         }
         Ok(output)