From e39096b7872816015975632484c58b7409d4d5d7 Mon Sep 17 00:00:00 2001
From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Sun, 22 Mar 2026 03:02:11 -0400
Subject: [PATCH] add tail_messages() for fast reverse transcript scanning

Reverse-scans the mmap'd transcript using JsonlBackwardIter,
collecting user/assistant messages up to a token budget, stopping
at the compaction boundary. Returns messages in chronological order.

resolve_conversation() now uses this instead of parsing the entire
file through extract_conversation + split_on_compaction.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---
 poc-memory/src/agents/defs.rs | 24 +++----------
 poc-memory/src/transcript.rs  | 67 +++++++++++++++++++++++++++++++++++
 2 files changed, 71 insertions(+), 20 deletions(-)

diff --git a/poc-memory/src/agents/defs.rs b/poc-memory/src/agents/defs.rs
index ab0dc82..d0f961d 100644
--- a/poc-memory/src/agents/defs.rs
+++ b/poc-memory/src/agents/defs.rs
@@ -455,19 +455,13 @@ fn resolve_conversation() -> String {
 
     let Some(path) = transcript else { return String::new() };
     let path_str = path.to_string_lossy();
-    let messages = match super::enrich::extract_conversation(&path_str) {
-        Ok(m) => m,
-        Err(_) => return String::new(),
-    };
 
-    // Take the last segment (post-compaction)
-    let segments = super::enrich::split_on_compaction(messages);
-    let Some(segment) = segments.last() else { return String::new() };
+    let messages = crate::transcript::tail_messages(&path_str, 25_000);
+    if messages.is_empty() { return String::new(); }
 
-    // Format and take the tail
     let cfg = crate::config::get();
     let mut text = String::new();
-    for (_, role, content, ts) in segment {
+    for (role, content, ts) in &messages {
         let name = if role == "user" { &cfg.user_name } else { &cfg.assistant_name };
         if !ts.is_empty() {
             text.push_str(&format!("**{}** {}: {}\n\n", name, &ts[..ts.len().min(19)], content));
@@ -475,17 +469,7 @@ fn resolve_conversation() -> String {
             text.push_str(&format!("**{}:** {}\n\n", name, content));
         }
     }
-
-    // Tail: keep last ~100K chars
-    const MAX_CHARS: usize = 100_000;
-    if text.len() > MAX_CHARS {
-        // Find a clean line break near the cut point
-        let start = text.len() - MAX_CHARS;
-        let start = text[start..].find('\n').map(|i| start + i + 1).unwrap_or(start);
-        text[start..].to_string()
-    } else {
-        text
-    }
+    text
 }
 
 /// Get recently surfaced memory keys for the current session.
diff --git a/poc-memory/src/transcript.rs b/poc-memory/src/transcript.rs
index 202140c..3fe3679 100644
--- a/poc-memory/src/transcript.rs
+++ b/poc-memory/src/transcript.rs
@@ -142,6 +142,73 @@ fn contains_bytes(haystack: &[u8], needle: &[u8]) -> bool {
     haystack.windows(needle.len()).any(|w| w == needle)
 }
 
+/// Reverse-scan a transcript file, collecting user/assistant messages
+/// until `max_tokens` tokens (~4 chars each) are accumulated. Stops at
+/// the last compaction boundary. Returns messages in chronological order.
+pub fn tail_messages(path: &str, max_tokens: usize) -> Vec<(String, String, String)> {
+    let (mmap, _file) = match mmap_transcript(path) {
+        Some(v) => v,
+        None => return Vec::new(),
+    };
+
+    let compaction_marker = b"This session is being continued";
+    let mut messages: Vec<(String, String, String)> = Vec::new();
+    let mut token_count = 0;
+
+    for obj_bytes in JsonlBackwardIter::new(&mmap) {
+        if token_count >= max_tokens { break; }
+
+        // Stop at compaction boundary
+        if contains_bytes(obj_bytes, compaction_marker) {
+            let obj: Value = match serde_json::from_slice(obj_bytes) {
+                Ok(v) => v,
+                Err(_) => continue,
+            };
+            if obj.get("type").and_then(|v| v.as_str()) == Some("user") {
+                if let Some(c) = obj.get("message")
+                    .and_then(|m| m.get("content"))
+                    .and_then(|c| c.as_str())
+                    && c.starts_with("This session is being continued") {
+                        break;
+                    }
+            }
+        }
+
+        let obj: Value = match serde_json::from_slice(obj_bytes) {
+            Ok(v) => v,
+            Err(_) => continue,
+        };
+
+        let msg_type = obj.get("type").and_then(|v| v.as_str()).unwrap_or("");
+        if msg_type != "user" && msg_type != "assistant" { continue; }
+
+        let msg = obj.get("message").unwrap_or(&obj);
+        let text = match msg.get("content") {
+            Some(Value::String(s)) => s.clone(),
+            Some(Value::Array(arr)) => {
+                arr.iter()
+                    .filter(|b| b.get("type").and_then(|v| v.as_str()) == Some("text"))
+                    .filter_map(|b| b.get("text").and_then(|v| v.as_str()))
+                    .collect::<Vec<_>>()
+                    .join(" ")
+            }
+            _ => continue,
+        };
+        if text.is_empty() { continue; }
+
+        let timestamp = obj.get("timestamp")
+            .and_then(|v| v.as_str())
+            .unwrap_or("")
+            .to_string();
+
+        token_count += text.len() / 4;
+        messages.push((msg_type.to_string(), text, timestamp));
+    }
+
+    messages.reverse();
+    messages
+}
+
 /// Get the timestamp of the compaction message at a given byte offset.
 /// Returns a human-readable datetime string, or None if unavailable.
 pub fn compaction_timestamp(path: &str, offset: u64) -> Option<String> {