Skip to content

Agent Sessions

v2.10 Breaking

This release includes breaking changes for platform teams planning a safe upgrade.

✓ No known CVEs patched
Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

agentic-workflow ai ai-coding-agents anthropic claude-code codex
+14 more
codex-cli copilot-cli cursor-agent developer-tools hermes-agent local-first macos openai openclaw opencode session-history session-management transcript-search usage-tracking

Summary

AI summary

Broad release touches Critical Fixes, Major Changes, Search, and Minor Changes.

Full changelog

Major Changes

  • Search: Unified Search is now Apple Notes-style: fast across all sessions, within a specific session, and compatible with all filters.
  • Search: Use SQLite full-text indexing to speed up global search after analytics indexing completes.
  • Search: Instant search now indexes full tool inputs and outputs for recent sessions (last 90 days), reducing the need to wait for background scanning.
  • Search: Tool output indexing now redacts embedded base64/data URL blobs to keep search responsive and the index compact.
  • Search: Instant search now uses token-prefix matching by default (for example, magic behaves like magic*) to improve identifier and structured-text recall without trigram/substr indexing.
  • Search: Background scanning of large tool outputs is now opt-in by default, keeping Instant search more responsive (Settings → Advanced).
  • Search: Instant indexing now samples long assistant messages and tool inputs (head + middle + tail) to reduce false negatives without indexing the full transcript.
  • Search: Instant indexing now captures more of long tool outputs (head + middle + tail slices) and keeps active Codex sessions searchable while they are updating.
  • Search: Multi-word Unified Search queries now behave like phrase searches (for example, exit code) to match transcript navigation and avoid accidental boolean parsing.
  • Search: Unified Search highlights now use token-phrase matching across punctuation/newlines, and Session view reports visible vs total match counts when role filters are active.
  • Search: Global search now accepts quoted repo: and path: filters, and background deep scans run at low priority with on-demand prewarming for opened sessions.
  • Search: The Search Sessions menu item and ⌥⌘F shortcut now reliably focus the global search field.
  • Search: Unified Search highlights matches in the selected transcript and jumps to the first match when switching sessions.
  • Indexing: Analytics refresh is now incremental (skips unchanged files and removes deleted ones) to reduce startup work and keep search data current.
  • Startup: When Codex sessions are already indexed, the app now shows the hydrated list immediately while scanning for newly created sessions in the background.
  • Transcript: Replace the read-only search box with a Unified Search navigation pill that appears only when Unified Search has a free-text query.
  • Transcript: Find in Transcript (⌘F) now opens a local find bar with its own query and navigation.
  • Transcript: Unified Search now auto-jumps to the first match after typing or switching sessions, and local Find uses a solid blue current-match marker aligned with the match.
  • Transcript: Session view now auto-scrolls to the last user prompt by default, with a Unified Window setting to choose first vs last user prompt.
  • Transcript: Color view is now called Session view and presents a colored timeline with improved formatting.
  • Transcript (Session view): Use system font for narrative blocks; keep tool call blocks monospaced.
  • Transcript (Session view): Blocks now render as rounded cards with subtle tints and left accent borders.
  • Transcript (Session view): User prompts now use a thicker accent rail, slightly brighter tint, and semibold first-line text for readability.
  • Transcript (Session view): Assistant blocks now use per-agent brand tints and the role legend matches transcript accents.
  • Transcript (Session view): Find highlights now mark matched substrings and add a line indicator for easier scanning.
  • UI: Consolidate unified-window status indicators into a single cockpit-style footer.
  • UI: In dark mode, the cockpit footer now uses a transparent HUD style with crisp borders instead of a solid fill.
  • UI: Menu bar usage now uses the same monospace, logo-forward cockpit styling as the footer.
  • UI: Persist split-view divider positions per layout mode (horizontal vs vertical) so switching layouts doesn’t reset pane sizes.
  • UI: Session list typography now uses monospaced text with softer timestamp gray, taller rows, lighter message counts, and footer-blue selection accents.
  • UI: Increase session list row height to 48px for easier scanning.
  • UI: Codex now uses a blue brand accent distinct from the cockpit footer tint; the transcript toolbar spacing is tighter, and the terminal view adds a subtle top border.
  • Onboarding: Replaced the onboarding flow with a four-slide tour covering sessions, agent enablement, workflow tips, and usage tracking.
  • Onboarding: Show all supported agents and discovered sessions in the tour regardless of Sessions toolbar filters; disabled agents appear as inactive.

Minor Changes

  • Onboarding: Refine tour icon and primary button colors to better match native macOS accents.
  • Sessions: Hide housekeeping-only sessions (no assistant output and no real prompt content) by default; use “Show housekeeping-only sessions” in Settings → General to reveal them.
  • Sessions: Show a live session count in the unified list that updates while indexing and when filters/search change.
  • Preferences: The Menu Bar pane now groups label options into sections and aligns toggles consistently.
  • Transcript: Removed the duplicate Jump to First Prompt control from the transcript toolbar.
  • Sessions list: The Size column can now be sorted.
  • Search: Tooltips now include keyboard shortcuts for search fields and navigation arrows.
  • Analytics: The By Agent card now auto-scales rows and falls back to an internal scroll when space is tight.
  • Menu Bar: Reset menu items now include weekday; menu bar label can hide reset indicators per provider.
  • Menu Bar: Removed pill backgrounds from the menu bar label for a cleaner, more native look.
  • Usage: Time-only reset strings now roll forward to the next day to avoid showing stale "<1m" countdowns.
  • Transcript: Toolbar controls now use monospaced typography to match the session list.
  • Sessions list: Added a context menu action to copy the session ID to the clipboard.
  • Sessions: Resume context menu actions now include the selected terminal app (for example, Terminal or iTerm2).

Critical Fixes

  • Search: Prevent missing results while the search index is still warming by falling back to legacy matching for unindexed sessions.
  • Search: Backfill missing per-session search data during incremental refresh so sessions don’t remain “not yet indexed” indefinitely.
  • Search: Claude sessions now keep transcript match highlights in sync with the active search query.
  • Transcript: Remove the ghost control inside the Unified Search navigation pill.
  • Transcript (Session view): Match counts now update when Unified Search is active.
  • Transcript (Session view): Search markers now appear only on matching wrapped lines and replace the block accent for that line.
  • Sessions: Auto-select the first session on launch so the transcript pane isn’t blank.
  • Sessions: Stabilize message counts for large sessions while full parsing completes to reduce list row jumps.
  • Copilot: Normalize tool output newlines when logs include escaped \\n sequences.
  • Parsing: Preserve non-zero exit codes in Gemini/OpenCode tool outputs and classify them as errors for error navigation.
  • Parsing: Droid stream-json now handles numeric timestamps, tool call IDs, and error flags in tool results.
  • Parsing: Treat Claude Code queue-operation and file-history-snapshot events as metadata so new versions don’t pollute transcripts or inflate message counts.
  • Parsing: Gemini sessions now account for model/tokens/thoughts fields in newer chat logs.
  • Parsing: Treat Copilot assistant.turn_start/end, tool.execution_start, and session.truncation events as metadata so they don’t clutter transcripts.
  • Transcript: Clearing Find now immediately clears match highlights.
  • Transcript: Terminal view no longer leaves a stale find highlight when the Find query is empty.
  • Transcript: Toolbar filters now use dot + count labels with compact navigation chevrons.
  • Transcript (Session view): Render Codex review blocks as labeled Review meta entries instead of user prompts.
  • Transcript (Session view): Split system reminder and local-command caveat blocks into meta lines so user prompts stay visible.
  • Transcript (Session view): Treat Claude local-command tag-only blocks as Local Command meta lines.
  • Transcript (Session view): Local Command meta blocks now render with a thin blue accent strip.
  • Transcript (Session view): Request interrupted-by-user markers now use a thin blue strip and regular text.
  • Parsing: Ignore empty JSONL lines during scanning for improved robustness.
  • Parsing: Surface Codex thread rollback events with readable text in timelines.
  • Claude: Avoid UI stalls when opening sessions with embedded base64 blobs (for example, Chrome MCP screenshots).
  • Claude Usage: Detect the Claude Code first-run terms prompt and surface a “Setup required” state instead of timing out silently.
  • Toolbar: Keep action icons visible and overflow actions accessible when a project filter is active.
  • Toolbar: Refined agent tabs and icon groups, updated the layout/theme controls, and tightened toolbar button sizing and hover states.
  • UI: Update Codex and Claude brand accents to blue and warm brown for clearer agent recognition.
  • Transcript: Rename view mode buttons to Session/Text/JSON, align them with HIG-style leading padding, and space the session ID control.
  • Menu Bar: When usage data is stale, reset indicators now show “n/a” instead of an incorrect countdown.
  • Claude Usage: Refresh usage automatically after wake when the usage strip or menu bar label is visible.
  • Menu Bar: Show an updating spinner next to reset indicators while probes run.
  • Search: Unified Search now accepts quoted repo/path filters with spaces.
  • Claude/Codex Usage: Add a conservative startup sweep for probe tmux servers and harden cleanup/timeouts to avoid orphaned CLI processes after stalled probes or restarts.

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Track Agent Sessions

Get notified when new releases ship.

Sign up free

About Agent Sessions

Local-first macOS app to browse, search, analyze, and resume supported AI coding-agent session history across Codex, Claude Code, OpenCode, Cursor Agent, Hermes, OpenClaw, Copilot CLI, and more.

All releases →

Related context

Beta — feedback welcome: [email protected]