Timeline fills itself from your own tools

The idea

The Timeline collector is a small program on a person's own computer. It reads GitHub and Jira through the sign-ins the person already has, with read calls only. It can show a preview that sends nothing, it sends only when the person runs a send or installs the service, and the same collector never records a copy of what it already sent.

The problem
Webhooks alone miss history and work done before setup.
Why it matters
Up to 30 days of past work and ongoing activity arrive with nobody typing a report.
Where it stands
In production as of Two production checks with operator test accounts; the collector read one repository of the person who ran them.

Crewshal team · · 4 min read

Recorded run · Production, operator test accounts · November 12–14, 2026

Can the feed fill itself?

A collector reads a person's own tools, shows a preview, sends on request, adds nothing on a repeat, and later collects on its own as a service.

Step 1 of 7Install

Test accountInstall and pair the collector

A test account installs the Timeline collector (version 0.5.14) from the served installer. The installed program matches the archive whose hash the check verified, and it lives in a temporary home folder, apart from the person's own runner. Pairing links it to the account's Timeline. Pairing sends no work records.

OutputInstalled

Program
Matches the verified archive
Home folder
Temporary, apart from the person's own runner
Work records sent by pairing
0

CheckRecorded steps

  • Install and pair (1 s)Passed
  • Set up the collector (0 s)Passed
  • Setup check (9 s)Passed

Where it ran. Two checks in production at crewshal.com with operator test accounts: November 12, 2026 at 01:18:37 UTC (the evening of the 11th in US Central) and November 14, 2026 at 04:31:27 UTC (the evening of the 13th). The second check erased its account afterwards. The collector read a repository of the person who ran the checks, through that person's existing GitHub sign-in; this page shows counts only.

The recorded files behind this replay
  • Production check: install, preview, send and repeat sha256 4509031c0ba0
  • Release note: Timeline signup and collector setup sha256 4248cc5ef3c5
  • Production check: the collector as a service, 21 checks sha256 da69add3a3d1
  • Release record: the service facts and the two script defects sha256 ada9394ffbe6
  • Release note: the collector service sha256 1fc05012dab4

Why it matters

Timeline is an automatic feed of what each person, outside agent, AI worker or automation did across connected systems. Nobody reports into it, and it needs no AI worker.

A webhook sees only what happens after someone sets it up. Work from the days before, and work in tools without webhooks, would be missing, and asking people to report it defeats the purpose. The collector brings that work in from the tools themselves, up to 30 days back, and the person decides when it starts to send.

How it works

  1. Install and pair. The person installs the collector from the served installer and pairs it with their Timeline. Pairing sends no work records.
  2. Your own sign-ins, read only. The collector uses the sign-ins that the person already has on that computer, and it makes read calls only. It reads a window of recent days: 7 by default, 30 at most.
  3. A preview you can run first. The preview command shows what the collector found and sends nothing. The setup puts it before the send command. A send reads the tools again; it does not send the preview.
  4. No copies. Each entry has a delivery key, so a repeat or a restart of one collector records nothing new. Since collector 0.5.16, a GitHub entry also names its event, so another collector's record of the same change shows as the same entry. See one event, one entry.
  5. On its own. Installed as a service (launchd on macOS; the code also writes a systemd service for Linux), the collector keeps running after the terminal closes and starts again at login. The service files hold no token, and a lasting refusal stops the service instead of a restart loop.

What the run showed

The replay above is two recorded checks in production with operator test accounts; the second check erased its account afterwards. The collector read one repository of the person who ran the checks, through that person's own GitHub sign-in. This page shows counts only.

  • On November 12, 2026 at 01:18 UTC (the evening of November 11 in US Central), the first check installed collector 0.5.14 from the served installer. The installed program matched the archive whose hash the check verified, and pairing sent no records.
  • The preview found 163 entries in the days it read, with no gap, and sent nothing. After an explicit send, Timeline held 163 new entries, each with the paired person as its reporter. A repeat added 0: all 163 were already there.
  • Sign-out removed the local credential. The check then revoked the connection and removed its temporary installation. It recorded no call to an AI model.
  • On November 14 at 04:31 UTC (the evening of November 13 in US Central), a second check installed collector 0.5.15 as a service on a Mac. With no command run by anyone, its first pass sent 107 entries. A forced restart and a scheduled pass 5 minutes later each sent 0 new entries and read again the 9 that were already there. Timeline still listed 107. Uninstall left no service and no process.
  • Two checks of that script failed on its own timing, not on the product. One looked for the new process 60 seconds after the restart, inside the service's 5-minute start limit. The other read the service list at the moment uninstall returned. Later reads showed the new process, and then no service and no process.

Limits

  • GitHub in these runs. Both runs read GitHub. The collector also reads Jira; these runs did not.
  • One person's repository. The checks read a repository of the person who ran them, not the work of a customer's team.
  • macOS only so far. The code also writes a Linux systemd service, but no check has run it on Linux.
  • Only while the person is logged in. A service collects only while the person is logged in on that computer and its sign-ins are valid. A collector that misses a run shows as late; one whose credential is revoked shows as stopped. See source health.
  • Copies across collectors. These runs used collectors 0.5.14 and 0.5.15, which keep copies out by delivery key only. A second collector's record of the same change shows as one entry from collector 0.5.16 on.

Frequently asked questions

Does the collector send my data without asking?

No work record leaves the computer until the person runs a send or installs the service. The setup shows a preview command first, and the preview sends nothing. A service, once installed, collects on its own.

Does running the collector twice create copies?

No. The same delivery key from the same collector records nothing new. In the runs, a repeat added 0 of 163 entries, and a restart and a scheduled pass added 0 to the 107.

Does Timeline need an AI worker for this?

No. The collector and Timeline work with no AI worker, and the first check recorded no call to an AI model.

  • Timeline: automatic cross-system activity feed

    An automatic feed of what one identity or Everyone did across GitHub, Jira, Slack and other connected systems; an identity can be a person, an outside agent, an AI worker or an automation.

  • Source health

    Each webhook, collector and token shows current, late, stopped, never or unscheduled from recorded facts, and a late source or partial read marks the feed incomplete.

    Recorded walkthrough

  • Own computer by default, one command

    A person connects their own computer with one pre-approved command; the official Claude Code and Codex CLIs run there under the person's own login, which the platform never opens.

Turn the numbers into shipped work.

Crewshal runs these choices for you: a persistent AI worker with memory and rules, on your Claude and Codex subscriptions.