Cogitator Planner, standalone copy
==================================

This runs the whole thing on your own laptop. No Docker, no install, no packages to
download. All it needs is Python 3.8 or newer, which most machines already have.

To start it
-----------
  Windows      double-click run-planner.cmd
  Mac, Linux   run  sh run-planner.sh

Either one starts the app on 127.0.0.1 and opens the board in a browser window.
Close the window and stop the script when you are done.

Screenshots
-----------
Paste one straight into a note or a record with Ctrl+V, or drop the file on the
window. It shows up as a thumbnail under the editor. To get it back out: drag the
thumbnail into a ticket, Explorer or Outlook and you drop a real png file, press Copy
and paste it wherever only a paste works, or press Save for a file on disk.

The sticky note
---------------
  Windows      double-click run-sticky.cmd
  Mac, Linux   run  sh run-sticky.sh

One note in a window the size of a real sticky note, with its own icon in the taskbar.
Keep it in front of everything with PowerToys Always on Top: Win+Ctrl+T.

It is the same note the board's notepad shows, saving as you type, and a screenshot
pastes straight into it. Run it on its own or alongside the board: both share one
engine and one planner.db, and whichever you start first starts that engine. Closing
the sticky note leaves the engine running so the next open is instant; it stops when
you log out, or close the small cogitator window if there is one.

The first run makes a .sticky-profile folder next to the script. That is the browser
profile that gives the sticky note its own window and icon. Delete it any time.

Dictation
---------
On Windows this already works. whisper-cli.exe, its libraries and an English model
came down in this folder with everything else, and are found on their own: start the
sticky note and the microphone button is there. Click it, talk, click it again, or
press Ctrl+Shift+D. Nothing was installed and nothing needs an administrator.

Every word is written down by this copy, on this machine. Nothing recorded is sent
anywhere, which is the point: a Comcast call can be dictated the same way a personal
note can.

On a Mac or Linux, or to use the Python engine instead, run setup-dictation.sh or
double-click setup-dictation.cmd. It installs into your own profile, still no
administrator rights, and fetches its own model once. After that a microphone button appears in
the sticky note and the notepad: click it, talk, click it again. Ctrl+Shift+D does
the same without the mouse.

The words are written down by this copy, on this machine. Nothing recorded here is
sent anywhere, which is the point: a Comcast call can be dictated the same way a
personal note can.

If the network blocks pip or the model download, run the setup on a machine that can
reach them, copy the downloaded model folder next to cogitator.py and rename it
whisper-model. It is found on its own. A whisper.cpp binary dropped in the folder
with its ggml model is picked up the same way, and needs no Python packages at all.

More accuracy on names and ticket numbers: take ggml-small.en.bin from the /client
page, put it in this folder and delete ggml-base.en.bin. Slower, noticeably sharper,
still faster than real time on a modern laptop. (With the Python engine instead, set
COGITATOR_WHISPER_MODEL=small.en before starting.)

The faction badges
------------------
They come down with this copy in marks.json and load themselves the first time you
start it. Delete that file before the first run if you would rather keep the plain
drawn glyphs. Clearing them later in the app stays cleared.

Where your data goes
--------------------
Everything lives in planner.db, in this same folder. It is a plain SQLite file. Copy
it, back it up, or delete it to start over. Nothing here talks to the internet and
nothing here calls an AI on its own.

This copy is separate from the one on the server. To move work between them, use
Extract on one side and Ingest on the other. The export file carries both the
knowledge base records and the task board.

Splitting a ticket
------------------
The split is six text rules applied in order, run locally in this file. It makes no
network call. The sheet tells you which rule fired and how many lines it matched.
