A daily summary of your outline, written by your agent

Once agents and other people write into your outline, it changes while you are busy with something else. Pando tells you who: the chip in the top bar shows who is writing, and Notify me when others write turns that into a notification. What they changed is in Changes, in the sidebar. What no screen keeps is the day itself. Changes covers your own outline, lists bullets newest first rather than by project, reaches back a week at most and writes nothing down, so what changed on a Tuesday has left it eight days later.

Yes, whenever it is asked: an agent connected to Pando calls the changes tool with the cursor it kept from the last run, gets back every bullet that changed since then with its text as it is now, and writes a short summary under a Daily bullet, whose note keeps the new cursor for next time. Pando does not start the agent, and the changes it reads do not say who made them or when.

Set it up

A connector you approve starts with your whole outline, and on Pando’s consent page you choose Read and write your outline or Read your outline only. The box beneath, ticked by default, gives it a bullet under Home to remember in, named after the client. Leave it ticked: that bullet is where the summary and its cursor can live at either level.

If you only want to look

Changes, in the sidebar, may be all you need. It opens on the last hour, with Day and Week one press away, and Day means the last 24 hours, not since midnight. Each row is one bullet with its time and a kind, completed, created, edited or deleted, and the kinds filter the list. Completed is exact: ticked inside that span and still ticked. A deleted row opens the Trash. It covers your own outline, not branches lent to you, and it summarises nothing.

Settings also has a Morning email: what today holds, at the hour you pick under Send at. That means what is under today, and every bullet not yet ticked anywhere that carries today’s date in its line or its note. It is a list of what is ahead, not of what changed, and Send one now sends it at once.

Tell it which branch to read

Asked with no branch named, the changes tool reads the agent’s memory bullet if it has one, not your notes, and a summary of that is mostly a summary of what the agent wrote itself. For your outline the call has to name it. The shared_with_me tool lists what the agent was given: for a connector, your whole outline, under the name of its top bullet, Home unless you renamed it. Approved to read only, it lists the memory bullet as well, because that bullet carries a grant of its own so the agent can write there. The root to send is the whole outline’s id, not the memory bullet’s.

The prompt

Write today’s summary of my outline in Pando.
- Under your memory bullet, find the bullet called Daily.
  Its note holds the cursor from the last run.
- Call changes with since set to that cursor, and root
  set to the entry shared_with_me calls Home, not your
  memory bullet; if it lists one entry, use that one.
  While more is true, call again with the cursor it
  hands back; if the cursor did not move, stop and
  tell me.
- Merge the pages by id, a later page over an earlier
  one. A bullet a later page lists as removed, or among
  the ids under a removed one, has left the branch.
- Leave out Daily and everything whose path runs
  through it.
- Under Daily, write one bullet: today’s date as
  YYYY-MM-DD on the line, the summary in its note. Group
  what changed by the branch it sits in, and list what
  was removed with how many bullets went with it.
- Replace Daily’s note with the last cursor.
If there is no Daily yet, make it, put the seq its
answer carries in its note, and write no summary today.

The date at the start of the line earns its ten characters. The Calendar in the sidebar lists, on each day, every bullet whose line or note carries that date, so each summary turns up on its own day. The Timeline beside it shows thirty days at once, starting today, and Thirty days earlier steps it back, which is the way to read the summaries one after another. A date quoted in the note counts too: a summary that repeats a line reading Dentist 2026-10-02 also turns up on 2 October, and in that day’s Morning email.

What comes back

• Each bullet that changed, once in each page, with its line as it is now, its note whole if it has one, and its path from the branch down. A bullet edited eleven times in one page is one entry, and within a page the last entry is the last write.
• A bullet counts as changed when its words, its note or its tick change, and when it is created, moved, re-ordered or shared. Folding and unfolding does not count.
• Each removed branch once, with the reason (deleted, moved out of the branch, or unlinked from it), how many bullets went with it, and their ids, up to 200.
• How many bullets outside the branch it asked about changed, as a number, without naming them.

What it does not say

• Who made a change. No entry names a person or an agent, and the agent’s own writes come back like anybody’s. That is why the prompt leaves Daily out: yesterday’s summary and cursor are in today’s answer.
• When. No entry carries a time, so since yesterday means since the cursor the last run kept.
• What a line said before. The text is the current one, and the old words are not in the answer.
• Whether a bullet was created, edited or ticked. A ticked todo comes back as its line. The agent can read which of those bullets are done now, and a tree read in outline mode marks them with d, but done now is not ticked since the last run: a todo ticked last month whose words changed today reads the same. For your own outline, the completed kind in Changes is the exact answer.

The first run, and a day missed

Asked with no cursor, changes reads the outline’s whole history from the beginning, a page at a time, and marks the answer resync: true, a fresh reading rather than a list of changes. It is the one call that costs more than reading the tree, and a summary written from it would list everything. So the first run writes none: it makes Daily and keeps as the first cursor the seq that write answers with, which is the same count the cursor keeps.

Nothing prunes the log, so a cursor kept on Tuesday still answers on Thursday. The summary then covers both days as one, since no entry says which day it changed. Each page reads at most 500 saves from the log, and more: true says there is another page to fetch before summarising.

Keep one cursor per outline. The numbers count per outline, so a branch that sits in somebody else’s outline needs a cursor of its own and a call with that branch as root. Pando refuses a cursor only when it is ahead of the outline it is sent to, so a wrong one that happens to be lower is answered, and answered wrong.

Where the summary can go

• Approved to read and write your outline: anywhere that is not protected, as long as the call names the bullet it writes under. The create tool refuses an agent’s new bullet that does not name its parent, which is why the prompt names Daily.
• Read and write, with the memory box unticked: the same, but there is no memory bullet to find Daily under, so the prompt’s first step has to name the bullet Daily sits under.
• Approved to read only: in its memory bullet and nowhere else. Daily under the memory bullet works at both levels.
• Read only, with the memory box unticked: nowhere, until you give it a bullet. The summary comes back as an answer in the chat, and the agent has nowhere to keep the cursor.
• Under a protected bullet: nowhere. A protected bullet refuses every change from an agent. A lock stops changes, not reading, so what changes in a protected branch still reaches the summary.

Holding it to one project

For a summary of one project and nothing else, move Daily into the project first. Then open the project’s bullet, open Agents and API keys in Settings, tap the agent’s row under Your agents, and press Let it reach only “…”, the bullet you are in. It then reads and writes that bullet and everything under it, remembers there, and reaches nothing else in your outline, even if you approved it read only. Its shared_with_me then lists that one bullet, so the prompt works unchanged.

Daily moves first because the press moves where the agent remembers to the project’s bullet and leaves the old memory bullet behind. A Daily left under the old memory bullet stays there, out of the agent’s reach, with its cursor and every summary so far, and the next run starts a new Daily among the project’s own bullets.

An agent you create under Agents and API keys with “Let it read and write this outline” and “Give it a bullet of its own to remember in” both unticked reaches nothing until you give it a bullet, which the same press does. Its key begins with pf_, is shown once, and goes in an Authorization: Bearer header. ChatGPT cannot send one. The Claude apps can only through the Request headers section of the Add custom connector dialog, which Anthropic’s docs call a beta available to a limited set of organizations.

Where this is not the answer

• A summary that is waiting without anybody asking. Pando starts no agent and sends an agent no notification, so the prompt has to be sent each time the summary is wanted. The Morning email is set to an hour you pick, but it lists what today holds, not what changed.
• A summary that has to say who did what. Neither the changes tool nor the Changes view shows who made a change, and the chip in the top bar shows only who is there at that moment, so a team’s stand-up needs the names from somewhere else.
• What a line said before the change. Only its current words come back.
• Work that happened outside your outline. The agent summarises what is in Pando and nothing else.

What it costs

Nothing, for 1,000 bullets, with every feature included and no payment card. A summary a day is 365 bullets a year with its lines in the note, more if they go in as children, and a summary you delete stops counting.

Deeper

• An agent that reads only what changed since last time
• Your weekly review: Claude prepares, you decide

Connect an agent, about two minutes · The twelve tools · Which note apps an agent can reach · Who runs this