Changelog

Most time tracking tools give you one input: a timer button. We think that’s backwards. Your work happens across screens, documents, conversations, and even handwritten notes — so time capture should meet you wherever the work is.

This week we shipped three new capture modalities:

The desktop agent runs a lightweight heartbeat process that groups application-level activity signals into time entries using configurable legal billing increments (6-min / 0.1hr rounding). No screen recording, no screenshots, no surveillance — just process-level metadata interpreted by AI. For attorneys and consultants who live in desktop applications, this closes the biggest gap in automated time tracking: the hours spent outside email and calendar.

Photo-to-time-entry via MMS — snap a picture of handwritten notes, a whiteboard, or a receipt and text it to TimeSentry. The image is processed through our multimodal vision model, which extracts structured time data (client, matter, duration, narrative) and creates a draft entry. This is the same inference stack that powers our email and calendar mappers, extended to unstructured visual input. For professionals who still take pen-and-paper notes in client meetings, this eliminates the transcription step entirely.

Word plugin document-change tracking — the Office add-in now monitors document mutation events and auto-saves time entries as you draft. Combined with the desktop heartbeat, this gives near-complete coverage of document-heavy workflows — brief drafting, contract review, memo writing — without any manual logging.

The AI agent’s tool-calling layer also gained access to your task and project graph, so conversational time management now resolves against real project structure rather than fuzzy name matching. The result: fewer corrections, faster approvals.

Two changes this week that matter most to firms with IT teams and custom workflows.

The desktop app now authenticates via a standard OAuth 2.0 flow with localhost redirect — no more copying tokens from the browser. This sounds small, but it’s the difference between “works for the founder” and “works for a 200-person firm with SSO policies.” Organizations that gate all software behind identity providers can now deploy the desktop agent without exceptions or workarounds.

API key lifecycle management is available in the settings panel. Generate keys, rotate them on a schedule, and scope them to specific integrations. This is the same auth layer that powers TimeSentry’s own agent integrations, now exposed so your team can build custom automations, connect internal tools, or pipe time data into your own reporting systems. We believe the best time tracking platform is one that doesn’t trap your data — it makes it programmable.

Agent companies get a new session ingest pipeline with relationship validation — the system verifies that an agent session maps to a valid company/user relationship before persisting any time data. This is the kind of invisible guardrail that prevents bad data from entering the system in the first place, rather than cleaning it up after the fact.

Time tracking shouldn’t require opening an app. The people who bill the most hours are the ones with the least time to log them — so we built a way to manage your timesheet through the tools you’re already using.

The TimeSentry AI Agent is a multi-channel conversational interface for time entry management. It ingests natural language over three transports and converts it to structured time data:

• Slack — DM the TimeSentry bot or @mention it in any channel. The agent resolves your identity via the Slack workspace mapping, queries your project graph, and creates or updates entries in real time. Ask “what did I bill to Acme this week?” and get an answer with a chart.
• Email — forward a client conversation or send a free-form note to your TimeSentry address. The agent extracts client, matter, duration, and narrative from the email body and thread context. This turns email — already the primary record of client work — into a zero-effort time capture channel.
• SMS/MMS — text your hours from anywhere. No app install required. Combined with the MMS vision pipeline, you can photograph handwritten notes and have them converted to entries on the spot.

The agent isn’t a chatbot bolted onto the side of a CRUD app. It’s a first-class integration type with tool-calling, chart generation, and access to your full project and client graph. The philosophy: if you can describe the work in a sentence, you shouldn’t have to fill out a form.

Also shipped: time entry merge and unmerge. Combine fragmented entries into a single line item for billing, or split a merged entry back out when you need granular detail. The merge preserves source metadata for audit trails — because simplifying a bill shouldn’t mean losing the receipt.

When you’re generating thousands of time entries per day across dozens of integration sources, even small edge cases compound. This week was a focused reliability sprint — the kind of work that doesn’t make flashy demos but determines whether you trust the system.

• Gmail dedup — eliminated duplicate entries caused by sent/inbox label overlap. Added a pre-commit dedup check to prevent race conditions during concurrent processing. Dedup boundary widened to ±2 days for all-day calendar events. The goal: you should never have to manually delete a duplicate entry.
• Email threading — thread-combining now collects all msg_ids for accurate dedup. Stale thread entries are replaced rather than skipped, and thread-based replacement only runs when use_email_threading is enabled. This matters for firms that bill per-conversation rather than per-email — one thread, one entry, correct every time.
• MS Graph calendar — reminder and to-do event types are now filtered at the sync layer so they never reach the mapper. Your timesheet should reflect work, not notification noise.
• Microsoft 365 token rotation — rotated refresh tokens are now persisted correctly, preventing silent auth failures. This was causing intermittent sync gaps for M365 users — the kind of bug where “it works most of the time” is worse than “it never works.”
• Timezone normalization — daily summary email queries and background hourly runs now use company-local time consistently, closing a UTC gap that appeared on date switchover. Time tracking is, at its core, a timezone problem — and we take that seriously.

Not every user wants a friendly dashboard with rounded corners. Some of our highest-volume users — managing partners reviewing firm-wide timesheets, operations leads auditing hundreds of entries — want density, speed, and zero visual clutter.

Terminal Mode is a Bloomberg-inspired interface theme built for exactly that. Amber text on dark backgrounds, area charts with gradient fills, crimson CTAs, and monospace typography where it counts. This isn’t a CSS skin — it changes chart rendering, color palettes, and layout density across every view. Toggle it on in Advanced Settings and your timesheet looks like a trading floor. Gated behind a company-level preference flag so firms can standardize the experience across their team.

Background task processing — bulk operations (remap, retry, archive) now run asynchronously with a real-time notification system. A notification bell surfaces progress as it happens. The UI applies optimistic updates on dispatch and reconciles when the task completes, so the interface never blocks on server-side processing. When you’re remapping 500 entries to a new project, you shouldn’t have to watch a spinner — you should be able to keep working and get notified when it’s done.

This is part of a broader principle: TimeSentry should feel instant regardless of the volume of data behind it.

Time tracking is the foundation, but the real value is what you build on top of it. This week we shipped two features that extend TimeSentry from a time capture tool into a financial planning surface.

FP&A scenarios — the financial planning module now supports multiple named scenarios (baseline, optimistic, conservative) with independent payment streams. Filter cash flow projections by scenario, drill down into line items with inline editing, and bulk-import data from Excel via a paste modal that parses tab-separated values into the scenario grid. For firm owners and CFOs, this means you can model “what if we lose this client” or “what if we hire two associates” directly against your real billable hour data — not in a disconnected spreadsheet.

Archive and spam filtering — time entries can now be archived individually or in bulk. The archive pipeline applies at the query layer, so archived entries disappear from the default view without being deleted. A spam classifier pre-filters low-signal entries (automated notifications, marketing emails) before they ever reach the AI mapper. The philosophy: your timesheet should only contain entries worth reviewing. Everything else should be handled automatically, with a clear audit trail if you ever need to look back.

Clio integration received no-charge field support directly on the time entry table — toggle no-charge status without leaving TimeSentry — plus a fix for a contact sync query that was degrading onboarding performance for large Clio accounts.

We built TimeSentry’s FP&A module because we saw firms exporting time data to Excel, building financial models manually, and then importing the results back into their practice management system. That’s three tools and a dozen steps for something that should be one surface.

The FP&A grid now implements spreadsheet-level interaction patterns: cell selection, fill-down, and paste-over ranges. Combined with inline editing and statement drill-down, you can build and adjust financial statements without switching tools. The grid handles cell-level focus management and keyboard navigation natively — it should feel like the spreadsheet you’re used to, but connected to live time and billing data.

Bulk retry and remap — when the AI maps entries to the wrong client or project (it happens — especially early in onboarding when the model is still learning your firm’s naming conventions), you can now select multiple entries, reassign the mapping, and reprocess them through the AI pipeline in a single operation. The remap runs asynchronously via the background task system with polling-based progress updates. This turns what used to be a tedious one-by-one correction into a single bulk action — and every correction makes the AI smarter for next time.

Time tracking only matters if it connects to billing. This week we shipped two new integration pipelines that close the gap between “hours worked” and “invoice sent.”

• QuickBooks Online — bidirectional invoice sync with date and project filtering. Client resolution runs through the existing entity-matching layer, so QBO contacts map to your TimeSentry client graph automatically. For firms that bill out of QuickBooks, this eliminates the manual step of recreating time entries as invoice line items.
• Notion — sync clients, projects, and contracts from Notion databases. Field mapping is configurable per-database, with external ID lookup for dedup and a preview panel that renders Notion properties in-context before you commit the sync. If your firm runs project management in Notion, your TimeSentry projects now stay in sync without manual upkeep.

AI explainability — every AI-generated time entry now exposes its reasoning chain. Click any entry to see which signals (email subject, calendar event, sender, project history) drove the mapping decision. A feedback loop lets you confirm or correct mappings, which feeds back into the model’s per-company learning layer.

This is a core design principle: AI should be auditable. If TimeSentry maps an email to the wrong client, you should be able to see why it made that decision — not just fix it and hope it doesn’t happen again. Explainability turns a black box into a system you can trust and train.

You can’t improve what you can’t measure. Most time tracking tools give you a list of entries and a total at the bottom. We think that’s a missed opportunity — your time data contains patterns, trends, and revenue insights that should be surfaced automatically.

New analytics dashboard with stacked area charts, trend lines, and revenue breakdowns computed from your time entry data. The backend resolves query bottlenecks in the revenue calculation path and caches results with automatic invalidation on time entry mutations — so analytics load instantly even on large datasets. See where your time goes, which clients drive the most revenue, and how your effective billing rate trends over time.

• Multi-axis filtering — filter the time entry table by hours, AI confidence score, billing status, and integration source. Filters compose and persist across sessions, so your preferred view is always waiting for you.
• Summary metrics bar — total hours, billable hours, and projected revenue rendered at the top of the timesheet with calendar week breakdowns. The numbers that matter, always visible.
• Calendar week view — a new visualization mode for seeing time distribution across the work week. Spot gaps, identify overloaded days, and understand your utilization at a glance.

The analytics engine is designed to scale with your firm. Whether you’re a solo practitioner reviewing a week of entries or a managing partner analyzing a quarter across 50 timekeepers, the experience should be the same: fast, clear, and actionable.

Time entries are the raw material. But professional services firms need to connect those hours to contracts, rates, staffing plans, and client relationships. This week we started building that connective tissue.

• Contracts & staffing — first-class data models with rate hierarchies, staffing assignments, and summary views. Contracts feed into the FP&A module and the billing rate resolver, so rate lookups now walk the chain: entry → category → contract → company default. This means the right rate is applied automatically based on who did the work, what type of work it was, and which contract governs it — no manual rate selection.
• Multi-Gmail support — connect multiple Gmail accounts to a single TimeSentry user. Each inbox is processed independently through the email pipeline with per-account dedup. For professionals who maintain separate inboxes for different clients or roles, all of that activity now flows into one unified timesheet.
• Chrome extension — capture time from any browser tab with AI project mapping. The extension calls the same AI classification endpoint as the email and calendar pipelines, so mapping is consistent across all input sources. Research in the browser, correspondence in email, meetings on the calendar — it all maps to the same project graph.

Onboarding was rebuilt with skippable steps, real-time Clio sync progress, and a folder-mapping wizard with auto-selection. Getting started with TimeSentry should take minutes, not hours — and it shouldn’t require reading a manual.

We started 2026 by rebuilding the infrastructure that lets us ship with confidence. The goal: deploy multiple times per week without breaking anything, and know immediately when something goes wrong.

• CI/CD pipeline — containerized builds with templated deploys to Azure Kubernetes. Stage and production environments are now fully isolated with unique revision suffixes per workflow run. Every pull request gets a staging deploy, every merge to main goes to production automatically.
• Observability — session replay with privacy-first masking, active company context on every error report, and user identity propagation across the frontend and API. When a customer reports an issue, we can see exactly what happened — without recording sensitive data.
• Automated release management — semantic versioning and changelog generation from conventional commits. 17 releases shipped in Q1 with zero manual tagging. This is why you see updates on this page nearly every week.

On the product side, we shipped cross-dimensional analytics matrix endpoints for reporting across client, project, and time dimensions simultaneously — laying the groundwork for the analytics dashboard that shipped in February.

We believe the speed at which a product improves is the most important feature. Infrastructure isn’t glamorous, but it’s what makes everything else on this page possible.