Skip to content

Agent Sessions

v4.6.1 Bugfix

This release fixes issues for SREs watching stability and regressions.

✓ 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

Bug fixes and stability improvements in Agent Sessions.

Changes in this release

Feature Low

Adds toolbar button to request a fresh usage reading per provider.

Adds toolbar button to request a fresh usage reading per provider.

Source: llm_adapter@2026-07-20

Confidence: medium

Feature Low

Allows reading Claude subscription usage by pasting a session cookie in Settings → Usage Tracking.

Allows reading Claude subscription usage by pasting a session cookie in Settings → Usage Tracking.

Source: llm_adapter@2026-07-20

Confidence: medium

Feature Low

Restores persistent right‑click hint for Quota Meter controls in On‑Demand mode.

Restores persistent right‑click hint for Quota Meter controls in On‑Demand mode.

Source: llm_adapter@2026-07-20

Confidence: medium

Bugfix Medium

Quota Meter no longer claims "no active session" when holding fresh data.

Quota Meter no longer claims "no active session" when holding fresh data.

Source: llm_adapter@2026-07-20

Confidence: medium

Bugfix Medium

Burn rate calculation no longer overstates subagent‑heavy sessions.

Burn rate calculation no longer overstates subagent‑heavy sessions.

Source: llm_adapter@2026-07-20

Confidence: medium

Bugfix Low

Web API no longer prompts signed‑in users to "sign in" when it cannot locate a session cookie on macOS 14/15.

Web API no longer prompts signed‑in users to "sign in" when it cannot locate a session cookie on macOS 14/15.

Source: llm_adapter@2026-07-20

Confidence: medium

Full changelog

What's New in 4.6.1

Highlights:

  • The Quota Meter stops claiming "no active session" while it's holding fresh data. Once your Claude CLI token lapsed, the calm "idle" verdict latched — and successful web-fallback reads could never clear it. The meter insisted nothing was running while the pasted-cookie web path served fresh usage every minute. Fresh live data now outranks the idle latch (a genuine auth problem still wins), Claude's idle tooltip becomes a recovery ladder — run the CLI, paste a cookie, probe — and web-served numbers are tagged via claude.ai so you can see where the reading came from.

Features:

  • Ask for a fresh usage reading from the toolbar. Forcing a hard probe used to be an undiscoverable double-click on the Quota Meter. It's now a visible toolbar button with a per-provider menu that tells you in the row itself when a probe is probing… or has failed, greys out providers it can't probe with a reason, and won't let two probes race from the toolbar, Preferences, and the usage strip at once. The old double-click gesture is gone.

Bug Fixes:

  • Burn rate no longer overstates subagent-heavy sessions. A session that keeps spawning subagents also keeps spawning fresh transcript paths, and each new path's first turn — a couple of seconds of cache-heavy setup — read as thousands of tokens per second. The existing session-level cap never fired for these sessions, because some path was always already measured, so orchestrated fleets showed burn rates up to 3.6x too high (1.6x median, validated against live ground truth). Provisional paths are now clamped to the session's best measured path rate, with pricing components scaled by the same factor — median error drops to roughly 1.25x.

Reminder: What You Got in 4.6:

  • Read your Claude subscription usage without the CLI — paste a session cookie. If you don't run the Claude CLI, Agent Sessions reads usage over the web instead. On macOS 14+ Safari stopped handing the claude.ai session to other apps, so that path had quietly become a dead end. Now you paste your claude.ai session cookie once in Settings → Usage Tracking: it's kept in your Keychain, needs no Full Disk Access, and sign-in never leaves your browser. A Test now button confirms it works on the spot.
  • Paste-a-cookie web usage for Claude. Settings → Usage Tracking accepts your claude.ai sessionKey (or the whole cookie header) and uses it directly for usage lookups — no Full Disk Access, no scraping. The app only holds the token you hand it, stored in the Keychain. Reading the cookie from Safari is kept as a legacy fallback for anyone it still works for.
  • The Web API path stops telling signed-in users to "sign in". When it couldn't find a claude.ai session it advised signing in "at claude.ai (default profile)" — impossible to satisfy on macOS 14/15, where Safari keeps the live cookie in a store apps can't read. The messages are now honest and cause-aware, and point to the fix that actually works: paste your session cookie.
  • The Quota Meter's right-click hint stays findable. In On-Demand mode the "Right-click for controls" hint used to show once and never again — but right-click is the only way back to the toolbar, so it now reappears on every hover.

Full release notes: https://github.com/jazzyalex/agent-sessions/releases/tag/v4.6.1

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]