# Show HN: Worklog, structured memory for Agent in one SQLite table

> Source: <https://github.com/xyb/worklog>
> Published: 2026-07-26 16:54:11+00:00

**🌐 English** · [中文](/xyb/worklog/blob/main/README.zh.md)

A local CLI worklog and task tracker backed by SQLite. Add tasks, log what you did, plan your day — all from the terminal, all in one file.

**Why wl:**

**Everything in one place.** Time (year → month → week → day) and projects (area → task) share the same tree and the same id space. No switching between apps.**A daily rhythm that sticks.** Set a goal in the morning, log as you go, recap at night.`wl day`

replays the whole day with stats.**Structured and queryable.** Every item carries status, priority, tags, props, and relations. Filter with`wl ls`

, track dependencies/splits with`wl relation`

(`block`

/`split`

/`related`

), search by meaning with`wl query`

.**AI-ready out of the box.** Every command outputs plain text or`-o json`

. There's a bundled[Claude Code skill](/xyb/worklog/blob/main/skills/worklog-cli/SKILL.md)so an AI assistant can drive it directly.**Local and transparent.** One SQLite file, no daemon, no account. You own the data.

```
pipx install pyworklog   # or: uv tool install pyworklog
wl init
```

Requires Python ≥ 3.9. Shell completion:

```
wl print-completion fish | source        # fish — add to config.fish
eval "$(wl print-completion bash)"       # bash — add to .bashrc
eval "$(wl print-completion zsh)"        # zsh  — add to .zshrc
wl add "write the report" -p A    # add a task, priority A
wl log 1 "drafted the intro"      # log progress → status auto-flips to DOING
wl done 1                         # mark done
wl day                            # replay today: plan + logs + stats
```

Start the day with a goal, log as you work, close with a recap. `wl day`

shows the full picture.

```
wl goal "ship the landing page today"     # morning intent (bare = read back)
wl log 42 "finished the hero section"     # progress note mid-day
wl recap "shipped it, copy still TBD"     # evening summary (bare = read back)
wl day                                    # the whole day: plan, logs, what changed
```

Real output from `wl demo`

(the tasks carry `-t work`

/ `-t personal`

— that tag is what sorts them into the day's buckets):

```
#9 2026-07-16 Thu · workday
  > 🎯 Send out the monthly report [1/1] ✅
     1. [x] #3 Write the report summary
  > 📝 Recap: Focus nailed - report's out, and I didn't chase the new idea. Weak spot:
    only scratched the AI tutorial. (written at 07-16 12:20)
  work
    ▸ planned
      [x] #3 [# ] Write the report summary
        · Draft written, sent to the team
  personal
    ▸ planned
      [ ] #4 [# ] Do the AI-agents tutorial

  ── 2026-07-16: 1/1 tasks with progress · DONE 1 · planned·not-done 1
```

`wl tree --by project`

groups the same data by project instead of by time:

```
▸ #1 [#A] Ship the monthly report  (1)
[x] [# ] #3 Write the report summary
▸ #2 [#A] Learn how AI agents work  (2)
[ ] [# ] #4 Do the AI-agents tutorial
[ ] [# ] #10 Build a tiny agent myself
```

Assign tasks to dates and move them through a status machine.

```
wl add "review PR" --parent 7             # add under a project
wl sched 42 tomorrow                      # plan for tomorrow; also: next-week, +3w
wl start 42                               # clock in
wl stop 42                                # clock out — elapsed recorded
wl defer 42 next-month                    # push to backlog; also: someday
wl done 42                                # close
```

Statuses: **TODO → DOING → DONE** · also **LATER** (backlog) · **WAIT** (blocked) · **CANCELED**

```
wl find "deploy"                           # full-text, hits highlighted
wl query "follow up with the client"       # finds related even if exact words differ
```

`wl query`

uses any OpenAI-compatible embedding server. For best results (LanceDB + CJK segmentation): `pip install 'pyworklog[semantic]'`

.

Every command accepts `-o json`

— place it before or after the verb.

```
wl -o json ls --para project              # array of open projects
wl show 42 -o json | jq '.logs'           # node detail, pipe to jq
wl -o json active                         # currently running tasks
# capture
wl add "task" -p A -t work                # new task, priority, tags
wl log 42 "what I did"                    # log progress
wl done 42                                # close; also: defer / cancel / wait

# plan & review
wl goal "deliver X by EOD"                # today's goal  (bare = read)
wl recap "shipped, one blocker left"      # end-of-day summary (bare = read)
wl day                                    # today: plan + logs + stats
wl sched 42 2026-07-01                    # schedule; also: tomorrow / +3w / someday

# navigate
wl show 42                                # full detail + timeline
wl ls --para project -t work              # list open work projects
wl tree                                   # whole structure, top-down
wl summary --week 2026-W25                # weekly aggregate by project

# search
wl find "keyword"                         # full-text search
wl query "follow up with the client"      # semantic + keyword hybrid
```

Every command has `wl <cmd> --help`

; `wl help <topic>`

browses topic docs inline.