Automatic time tracking for Jira: what "automatic" actually means, tool by tool
Every Jira time tracker calls itself automatic. Which of the four jobs in a worklog each one really automates, checked against vendor docs in October 2026.
Contents
Automatic time tracking for Jira records your work and turns it into Jira worklogs, and in practice every vendor automates a different part of that. A worklog is four jobs: noticing that time was spent (capture), deciding which issue it belongs to (attribution), deciding whether it counts as work at all (judgment), and writing it into Jira (delivery). Capture and delivery automate well. Attribution automates partly. Judgment should not be automated, and we don't.
I build Planim, an automatic time tracker that records your day on your own machine and matches it to Jira issues with models that run locally. Facts about the other tools below come from their own pricing and help pages, checked in October 2026.
The four jobs
| Job | What it is | Automates? |
|---|---|---|
| Capture | Recording that 09:12 to 10:40 went to ribbon.ts in VS Code and three MDN tabs | Yes, completely |
| Attribution | Deciding those 88 minutes belong on ATT-142 | Partly |
| Judgment | Deciding the 20 minutes in Slack were work on ATT-142 and not a tangent | No |
| Delivery | Writing a worklog with a start time, a duration in seconds and a comment into Jira | Yes, completely |
Capture and delivery are solved problems. An operating system will tell a program which window has focus, what its title is, and when the user stopped typing; git knows the branch. At the other end, the Jira REST API takes an issue key, a start timestamp and a number of seconds and answers 201. Neither needs a model.
Attribution is where products differ. A window title that contains ATT-142 is easy. The pull request you reviewed in six tabs, the incident that touched four issues, the hour of thinking in a notebook: those are hard, and the question to ask of any tracker is what it does in the second case.
Judgment is a team policy. Was the Slack thread work on the issue? Does the interview go on a worklog? Teams answer differently, and the same team answers differently the week before an invoice. A model can propose; a person commits, because the worklog is a shared record that feeds reports, invoices and sometimes payroll.
Which job each tool automates
Checked against each vendor's own pages in October 2026. "Writes worklogs" means the tool itself creates a worklog on the Jira issue, as opposed to importing issues as projects.
| Tool | Capture | Attribution | Judgment | Writes worklogs to Jira |
|---|---|---|---|---|
| Tempo Timesheets | No device capture. Suggestions from your calendar, Jira activity, and the VS Code and JetBrains extensions if installed | You pick the issue | You, one click per suggestion | Yes, natively |
| Timely | Memory records apps, websites and some file names, uploaded to Timely's cloud | AI drafts entries for your Jira activity | You approve each draft | Yes, sync on Premium and above |
| Clockify | Auto tracker records apps and sites, kept locally for 45 days | Manual: you turn recorded activity into entries | You | Only through the Jira plugin timer; auto-tracked time stays in Clockify |
| Toggl Track | Timeline records apps and sites | Manual | You | No. The Jira sync is one-way, Jira into Toggl |
| Memtime | Records apps, documents and sites, stored offline on your machine | Rules you write by program, URL or window title; the rest by hand | You | Yes, on the Connect plan and above |
| Rize | Records apps, sites and window titles, filtered on the device, uploaded to Rize's cloud | Automatic categories and projects | You review entries in its panel | Through a Zapier flow you build |
| Planim | Focused app, window title, browser tabs on macOS, idle, lock, file edits, git branch. Stored on your machine | Ten on-device channels vote, with evidence shown | You, one key per draft | Yes, confirmed drafts only |
Three patterns fall out of the table.
Capture, then sorting. Clockify and Toggl record your day faithfully and hand you the list; Memtime adds rules you write. The remembering is automated. The assigning is still yours, one block at a time, dozens of decisions on a fragmented day.
Attribution from a narrow window. Tempo sees your calendar, what you touched in Jira, and file saves in VS Code or JetBrains if you install its extensions. It cannot see the browser, the terminal, Figma, or any other app. Suggestions are good for meetings and thin for everything else.
Attribution in the cloud. Timely attributes, and to do it uploads the activity log. For some people that is fine. For anyone under a client NDA, a works council, or a plain preference not to send window titles to a third party, it is the reason they still log time by hand.
What we automate, and where you stay
The part of Planim I care most about is the inspector. Every proposal shows its votes: which channel said what, with what score, and the text it saw. When the app is wrong you can see why in two seconds, and your correction goes into a log that retrains the channel weights once there are fifty verdicts. Each retrain is replayed against the whole log and kept only if accuracy and coverage did not drop.
Behind the inspector, capture is local: the focused app and window title once a second, browser tabs on macOS, idle after three minutes without input, screen lock, file edits and the current git branch.
Attribution is a vote. Ten channels look at each episode. Three are anchors: an issue key in the title or URL, a known URL pattern (Jira, GitHub, Figma, Notion, Asana, Trello, Google Docs), a file path. Two are small models running on the CPU: an embedding of the episode's text compared against your open issues, and a reranker over the top candidates. Three are your history: the nearest past verdicts, which issues tend to appear together, and aliases the app learned after you confirmed the same match twice. The last two use priors and what overlapping episodes in other streams say, so an edit made while a call window had focus keeps its own issue. The votes are fused by rank, anchors count triple, and the result is a proposal with a calibrated confidence.
Judgment stays with you, and the interface makes it one key. Enter confirms. R reassigns. I marks it Not work. N jumps to the next draft that needs attention, which is any proposal under 80 percent confidence. Confident proposals wait too; nothing is pushed until you press Push worklog, and then it is one POST per confirmed draft.
Five questions to ask any vendor
- Where does the activity log live? On the machine, or on their servers? The answer decides whether the tool can be installed under some contracts.
- Does it create the worklog in Jira, or a time entry in its own database? Ask for the Jira screen showing the worklog.
- How does it pick the issue when no key is in the window title? If the answer is "you pick", the tool only records.
- Can I see why it picked that issue? Without the evidence you can override a match but never teach it.
- What does it do when it is not sure? It should flag the draft and ask.
Which one to pick
If your day is mostly meetings and Jira, Tempo's suggestions already cover it. If you want a complete local record and do not mind assigning it yourself, Memtime does that offline. If you want the assigning done for you and the upload is acceptable, Timely is the finished product.
Planim is the fourth option: the issue picked on your machine, the evidence shown, and only confirmed worklogs in Jira. It is being built for macOS 14 and Windows 10/11; the early-access list below is how to get it first. The same tools, one by one, are in the best automatic time trackers for Jira.