Brings Pulse up to date with this autumn's Claude Code and Claude Desktop changes. - Your costs will read about 10% higher. Pulse was counting each response's first transcript line, which Claude Code writes while the response is still streaming, so about a third of responses were recorded with only part of their output. Every response is now counted from its complete line, and history is recounted wherever Claude Code still keeps the transcripts. - Cost now counts advisor calls and both attempts of a model fallback, each at its own model's rates. - Haiku 5.5 is priced by prompt size, Sonnet 5.5 cache reads at $0.10, and US-only inference and web searches are included. - Resumed and forked sessions are no longer counted twice. - Effort is set per model, as Claude Code now saves it. The old single setting had no effect on Opus 5.5 and newer. - The model picker offers the current models, and renamed sessions show your name. - Grant Access for symlinked skills now asks for the folder the links actually point into. - MCP servers are read from every scope Claude Code uses, editing keeps a server's transport, and the connection test is more accurate. - Your history moves to a new format on first launch. It's backed up first, and older days are re-priced at current rates.

A follow-up to 1.8.1 that fixes what a review of that release found. - Long-time users' oldest history is labelled as estimated again. 1.8.1 wrongly marked it as measured, showing months of "0 lines". - Opus 5.5 is priced at $4/$20 per million tokens ($8/$40 in fast mode) and Sonnet 5.5 at $2/$10. Recent saved history is re-priced automatically. - Newer model versions take the price of the nearest version Pulse knows, not their family's oldest rate. - The live usage gauge now reads only what changed since its last refresh, so it uses far less CPU and disk while Claude works. The figures are unchanged. - The gauge no longer wakes up for Claude Desktop's caches and logs.

A correctness release. Every fix below has a regression test behind it. Your history stays put Connecting or disconnecting Claude Desktop, or upgrading across a schema change, could silently drop days whose transcripts Claude Code had already pruned. Saved days are now kept whenever a rescan can't reproduce them. Costs that mean what they say - Today, Week and 30 Days are cut at the time of each response, so a session that ran past midnight no longer carries yesterday's spend into today. - Fast-mode responses keep their premium all the way to the dashboard cards, which now agree with the usage gauge. - Sonnet 5 and Fable 5.1 are priced at their own rates rather than their family's. - Recent history can be re-priced when a rate is corrected, instead of keeping a wrong figure forever. Measured means all of it "Measured" lines and active time now mean every day in the period was measured. A partly covered period blends measured days with the estimate for the rest and keeps the "(est.)" label. A fully scanned day with no edits reports zero rather than an estimate. The usage gauge is live It refreshes on its own a few seconds after Claude activity settles, re-checks your notification thresholds, and shows when it last updated. MCP connection test Rejects error replies instead of reporting "Connected", accepts servers that answer with an event stream, sends the headers configured for the server, and reports an authentication failure as such. Also fixed Line counts for created files ending in a newline or using Windows line endings, and a rejected first write that used to leave the unverified file behind.

Measured, not estimated Active time and lines of code used to be estimates. Claude Code records the duration of every turn and the diff of every edit — Corixa Pulse now counts them. The old active-time guess discarded any pause over ten minutes as a break, throwing away exactly the long unattended runs that agentic work is made of; the old line count ran at about a third of the real figure. Anything still estimated is labelled as such. What each effort level costs The Cost & Tokens section now splits your spend by reasoning effort — responses, total, and cost per response for each level. Settings could already set the default effort; now the dashboard tells you what that choice is worth. Skills, Servers & Subagents A new dashboard section shows what actually did the work: which skills ran, which MCP servers were called, which subagents were spawned, each with a response count and cost. It reads your transcripts rather than your config, so it includes claude.ai connectors that appear in no local config file. The Skills, Agents and MCP Servers lists now carry 30-day usage badges. Sessions have real names again Claude Code stopped writing the session index the app used for summaries, so sessions fell back to a truncated first prompt — sometimes raw command markup. The app now reads the generated session title from the transcript. Fast mode is priced correctly Fast mode bills Opus at double the standard rate. The app wasn't reading the speed the transcript reported, so every fast-mode window was costed at half. Cost, Usage Intensity, the menu-bar gauge and threshold notifications now price it correctly. Also in this release: local placeholder responses no longer show up as a phantom Sonnet entry in the cost breakdown; CSV export gained Active Minutes, Lines Added, Lines Removed and Files Created; and your statistics history rebuilds once on first launch, carrying older days across untouched.

Optimize for better approximation of usage and spending of usage limits of subscriptions.

Adjustment of usage calculations and limits according to latest changes in Claude

• Usage Intensity now tracks message volume — the dimension that matches how Claude.ai actually meters your session — so a single high-output session no longer shows a misleading 100%+ reading. The API-equivalent dollar cost stays visible as a reference figure. • More accurate cost estimates: 1-hour prompt-cache writes are now billed at their real rate. • Sessions launched from temp directories are correctly re-attributed to their real project. • Reliability and polish throughout.

Added Sharing of achievements and progress

Add visibility to tools that belong to plugins, show marketplaces.

Fixed MCP server visibility issues

Renaming Projects and Project history Archiving Projects/history

Removed MCP Test connection due to technical issues, improved MCP editing.

Smoother Activity view Switching between tabs no longer resets the Activity panel. Your previous data stays visible while refreshing in the background — no more blank screen flicker. The selected date is also remembered when you return. Antigravity editor support Projects can now be opened directly in Antigravity from the project detail quick actions. MCP Server Editing Full MCP server management in-app — no more hand-editing ~/.mcp.json. Add/Edit in modal form (stdio or HTTP, args, env vars, live validation), Delete with confirmation, and Test Connection for any server (HTTP responses inline, stdio opens Terminal). Atomic writes with automatic backups, and unknown fields are preserved. Off by default — enable via Settings → App Features → MCP Server Editing. Honest Usage Intensity "Plan Utilization" renamed to Usage Intensity. Dropped the hardcoded token estimates that misaligned with Claude.ai's real rate limits; now only uses dimensions matching actual plan limits, with an estimate caption.

Fixed visibility of MCP servers

See everything Claude Code can do Commands, agents, and MCP servers from your installed plugins now appear alongside user-level tools — not just skills. Each list is grouped by source so you can see exactly what comes from which plugin. Claude Desktop integration Stats from Claude Desktop agent-mode sessions are now included in your dashboard. Connect your Desktop data directory once and historical sessions merge automatically. Skill companion files Skill detail view now shows companion files — references, agents, and assets — in a structured panel, so you can see everything a skill brings along. Automatic rollup backups Your stats history is now backed up daily before any changes, with the last 5 snapshots kept automatically. Peace of mind for irreplaceable data. Only enabled plugins Disabled or duplicate plugin installs no longer clutter your tool lists.

Symlinked Skills Support Skills symlinked from ~/.agents/ (e.g. installed via npx skills) now appear in the skills list. A one-time access grant is all that's needed.

Historical Statistics Rollup Anthropic retains your Claude conversation data for just 30 days (7 days for API usage). Until now, Corixa Pulse mirrored that limitation — statistics only covered the most recent 30-day window. Version 1.2 changes that. Usage data is now persisted locally, surviving well beyond Anthropic's retention period. On first launch after updating, the app performs a full scan of all local Claude Code JSONL session files — including subagent activity — and builds a comprehensive historical statistics file. This one-time scan aggregates messages, sessions, tool calls, token usage by model, hourly activity distribution, and per-tool breakdowns for every day of available data. Subsequent launches update the statistics incrementally, scanning only new session data since the last update. Startup remains fast regardless of how much history you have accumulated. Monthly History Cards Three compact summary cards are now pinned to the bottom of the sidebar, providing an always-visible snapshot of your usage trajectory. Each card displays a per-month summary including: * Messages sent during the month * API-equivalent cost estimate * Sessions started * Tool calls made * Trend percentage compared to the previous month (up or down) * Daily activity sparkline showing usage distribution across the month The current month's trend is pro-rated based on elapsed days, so it reflects your actual pace rather than comparing an incomplete month against a full one. A "pace" label indicates when the trend is projected. Use the arrow navigation buttons to page through historical months and compare your activity over time. Statistics Backup & Restore Your historical statistics can now be exported and imported from the Support page. * Export saves your complete usage statistics to a file you can store anywhere. * Import restores previously exported statistics, preserving your history across reinstalls, machine migrations, or transfers to a new Mac. Your data is yours to keep — even if you need to start fresh with the app. Extended Period Filtering The Year and All Time period filters now display genuine historical data spanning all available months, rather than being capped at 30 days. * Year shows a full 12-month rolling window drawn from your persisted history. * All Time aggregates every month of data ever recorded, giving you a complete picture of your Claude Code usage from the earliest session file on your Mac through today. Compare your current week against your entire usage history to spot trends and understand how your workflow has evolved.

Cost & Token Analytics See how much value your Claude subscription delivers. The new Cost & Tokens dashboard shows your API-equivalent usage value with ROI context, output/input token totals, and cache hit rate — all parsed from your local session data with zero configuration. Per-Model Breakdown Understand your model mix at a glance. See cost distribution across Opus, Sonnet, and Haiku with proportion bars and percentage splits. Cache Efficiency Visualize how prompt caching saves you money, with a stacked efficiency bar and dollar savings estimate. Plan Utilization Track your 5-hour window usage history over the last 30 days. See how often you approach your plan limits with color-coded utilization bars. Plan Selector Choose your Claude plan (Pro, Max 5x, Max 20x) in Settings. The app prompts you on first launch and tailors all utilization metrics to your actual limits. Open Projects in Your Editor Right-click any project to open it directly in Cursor, VS Code, Windsurf, or Warp. Open CLAUDE.md and MEMORY.md files in your editor with one click. Period Filtering Cost, token, and model breakdown data now follows the dashboard period filter — switch between Week, Month, Year, and All Time.