​

Ponder

​

Sources

Docs: Pondering Web App: Open Ponder from your Workspace sidebar; each Workspace has its own address. Ponder is where people and AI work through a Problem together. The Problem keeps the ordered Message thread; each ordinary response uses the Persona, profile, and Powers selected for that request. An explicit Process uses a separate Run; Processes are not available yet. Another person can return to the thread without reconstructing it from a temporary chat.

​

Set appearance and sound

The Workspace sidebar places appearance and sound controls immediately before Parable Pets. Select the appearance icon to cycle Auto, Light, and Dark; a toast confirms the selected mode. Auto continues to follow the operating system. Sound effects are on by default. Select the sound icon to turn them off or back on for this browser. The choice applies across Parable and does not enable new sounds on surfaces that do not already use them. Audio assets remain unloaded until an existing interaction requests a sound.

​

Start with a Problem

If no active Parables are available to you and you can create one, Ponder creates and opens Your First Parable. Other people’s private Parables do not prevent you from getting your own. Reopening Ponder uses your existing Parable. A link that names a Parable opens that Parable even when the sidebar list has not caught up yet; Ponder re-reads the list once before it treats the link as stale and opens your first Parable instead. When Home is the only Page, Ponder starts at full width. The left/right arrow control restores the Page and preview alongside it; the up/down arrow control expands or restores the conversation’s height. These controls work independently from every tab and remember the width choice. The composer stays centered at a comfortable width. An empty conversation shows the Ponder orb without introductory copy. Ponder expanded in both directions shows the Parable title in Editorial New, at the same size as the breadcrumb title. Side-by-side, its header says Ponder. The restored view keeps that title as the first breadcrumb, followed by compact uppercase Piece names. Expand and restore animate smoothly, respect reduced motion, and play transition sounds only when your sound preference is enabled. The Page’s bottom editor starts closed. Choose a document tab to open it; Ponder remembers whether you left that editor open or closed. A Page without Panes shows the dotted canvas alone. Handle-display menus and package-transfer controls are temporarily hidden. The mode dropdown stays visible while your Parable is Personal. Choose Publishing there to save Versions and invite participants when you have permission to change its Collaboration Mode. A Problem is the shared thread for an outcome, question, risk, or decision. Give it a name that a teammate can recognize later, then keep working in the same thread as the surrounding Parable changes. To create one from the Problem picker, type its name in search and choose Ponder Problem. The new Problem keeps the name you typed. Use a separate Problem when the work has a different outcome, owner, or review path. Keep the same Problem when you are refining the same outcome across several messages or pieces of a Parable.

​

Give Ponder the right context

Ponder can work from context already available to the Problem and can use authorized tools to inspect the Workspace when a request needs more. A Problem keeps its connection to its Parable even when no specific Piece is selected. When the Parable changes, its attached context still identifies the Parable; Ponder can read current details before acting. Merely opening another Parable, change, or Piece does not silently rewrite the Problem before your next message. Each Message receives the same definition revision and Latest or pinned Refresh reading used by the open Parable’s React and Plots, together with the selected Changes and Piece. Queued Messages and approval continuations retain that view. This applies to every tool in chat and Processes: context links and the current view are optional. Name another Parable, Piece, or set of Changes to work there; Ponder uses that target instead of replacing it with the open view. Without a view, Ponder can discover an authorized target and asks only when the choice remains ambiguous. A target, required fields, or current revision may still be needed by the action itself. Tools that manage context links naturally need the link they act on. Context helps Ponder understand the work; it does not give Ponder new access. You and Ponder can use only the information and actions allowed by your current Workspace access.

​

Choose how Ponder should approach the work

The composer starts with Automatic under Problem, which is the current Parable’s primary Problem; open the list to work in a specific Problem instead. Every response also has a Persona: a reusable approach such as coordinating, composing, checking, or bringing several perspectives together. Automatic under Persona is the Coordinator; Ponder resolves an active Persona before each response. Choose Simple, Standard, or Strong in the input footer. Each is a saved ranked-provider configuration; new Personas start with the Workspace’s default chat models, which are Portal unless the Workspace sets its own. The names do not add hidden prompts or change model effort by themselves. The Profile settings gear beside these choices is behind the ponder_profile_settings Datadog feature flag. It is off by default, and off in local development until you switch it on from the dev panel’s Flags tab; while it is off, no private copy changes what a send asks for. When it is on, the gear opens the current profile’s ranked models in a dialog. Save for Me keeps your edits as a private copy that applies only to you; Update Persona shares them with everyone who uses the Persona, after a permission- and revision-checked save; Reset discards a private copy. See Personas for ranking and native configuration. The input footer holds four groups spread across it: attachments; Simple, Standard, and Strong; Persona and Powers; and the people in the Problem beside Send. Persona and Powers show their small uppercase label and an icon, not their value: the Persona’s glyph (masks while Automatic), and for Powers the chosen Power’s glyph or a count when several are chosen. Point at either to read the names. One row above the input holds Participation and Problem, each with a small uppercase label before its value; Automatic uses the muted text treatment. Release and Refresh sit at the Parable’s top right. Expand a ranked model row to open its complete Configuration, including structured output and other native operation settings. Unset values are shown as placeholders and do not change the saved settings. Saving those settings does not establish that the selected provider can execute every operation; unavailable adapters and privacy requirements still apply. You can also select one or more Powers when the work needs a reusable set of instructions. The Powers control opens a searchable list; each row carries a checkbox on the left that reflects whether the Power is selected, and choosing the row toggles it. Hover a row to reveal its Edit pencil. Personas and Powers guide the approach; neither can expand your permissions. Each Message captures your open Parable, Changes, version, and Piece even when none is attached to the Problem. Queued Messages and approval continuations keep that view. Context links are optional reference material, not permission to use a tool; Ponder can discover or ask for a target when no view is selected.

​

Send one clear request

Write what you want to achieve, what a good result looks like, and any boundary that must remain true. Use the arrow to send the message and begin the work. While Ponder is answering, the composer stays open. Sending again queues the Message: it waits in a stack above the composer with your portrait and its place in line, and goes out on its own when the current response settles. Remove a queued Message from the stack if you change your mind; Stop, beside Send, ends the current response. Only your own queue is shown; other people’s queued Messages are not yet visible. Problem, Persona, and Powers remain available while Ponder answers. A queued Message keeps the Persona, Powers, attachments, and context selected when you sent it; later changes apply to later Messages. Switching Problems leaves the other response running. Stop returns your queued Messages to the composer, above your draft. After a failed response, use Resume to continue a held queue. To redirect your own reply while it runs, press Cmd+Enter instead. Your words wait in the stack marked “steering this reply” until Ponder takes them, after the step it is on and before it thinks again, then appear inside the reply as Steered. You can steer only a reply to your own Message; attachments are sent as a queued Message instead, and words steered after the reply ends become your next Message. The composer’s settings menu, the gear beside the attachment control, holds Thinking (Simple, Standard or Strong for the next Message), Clean up text and Keep the Parable named. Clean up text fixes spelling, spacing, capitalization and closing punctuation before a Message is sent. It replaces only words it lists as corrections, using the Workspace’s names and the Parable’s terms, and never adds, removes or reorders ideas. It is on until you turn it off, and the choice is remembered in this browser. A cleaned Message shows the broom in its footer; hover or focus it to read what was typed. Everyone on the Problem sees it, as they see the Message. While a send waits on it, a broom sweeps beside the gear. If cleanup fails, is busy or takes longer than a moment and a half, or you switch it off while it runs, the Message goes out as you typed it. If the Workspace’s AI policy does not allow cleanup, the item is greyed out and says why, for the rest of the session. Steering words are sent as typed. Keep the Parable named (Summarize) lets Ponder keep the Parable’s name, description and icon up to date from the Messages sent while viewing it. It is one setting for everyone on the Parable, not a personal one. A new Parable starts with it on, so the first Message names it. Editing the name, description or icon yourself turns it off; anyone who can edit the Parable can turn it back on, and everyone else sees its state, greyed out. Stop also appears for a reply that is still running after you reload or return to the Problem, and ends it for everyone in the thread. A reply paused on an approval is waiting for you, not running, so it offers the approval instead of Stop. The thread follows new content smoothly while you are at the bottom. Scrolling up keeps your reading position until you return to the latest Message; reduced motion uses immediate scrolling. Sending a Message or answering a question returns you to the latest Message, and the reply grows in place beneath it without redrawing the turn you just sent. Ordinary responses stream into the thread as the model writes. Each reply is signed Ponder followed by the Persona that answered, for example Ponder Coordinator. A reply’s footer shows its time. People who can debug Ponder also see a Debug icon on each reply. Its view opens with the reply’s reported AI usage and model request count: a tool loop updates the amount after each model step whose charge is known, marked so far, and it settles on the response total when the reply finishes. A partial receipt is marked recorded; a finished reply whose provider did not report a charge reads Unavailable instead of zero. Below the usage are the reply’s status, identifiers, times, and tool steps, with Copy details and, for a Run, Run diagnostics, which lists the Run’s safe events step by step. Ponder answers in a bubble of its own, one shade lighter than the thread in both themes, beside a round disc that shows the answering Persona’s icon. Bubbles are capped at 80% of the column, except in a narrow column, where they take its full width. While Ponder works, the disc shows an animated orb in place of the icon, and the row at the head of the reply reads Thinking… or the name of the tool being run. The row stays live until the answer itself starts streaming in, even when Ponder wrote a few words before its next step, and goes when the answer lands. Ponder’s chain of thought sits behind a brain beside the Persona’s name, shown when you point at the reply, with THOUGHT FOR 4 SECONDS beside it once the answer lands (or the number of steps, for a reply you open later). Turn it on and the thought opens there, as its own bubble between the name and the answer: each tool step with its outcome (failed and disallowed steps are marked), opening again to its parameters and result, and the model’s reasoning in the muted voice. Opened before Ponder’s first step, it shows Gathering context… while Ponder reads the thread. Turn the brain off to hide it again. A reply with nothing to trace shows no brain. A Pane preview or a Ping question a tool produces stays in the answer, not behind the trace. The question card appears only in the reply that asked it; the thread does not list Pings, Possibilities, or Proofs after the latest Message. A single question is the card’s title. Each option is a row with its name in capitals and its summary beneath. A question whose options carry reasons shows Pros and Cons as bulleted columns beside them; a question without reasons keeps plain rows. Choose an option to answer, then use the note icon on its row to add a note or a picture to that choice. The icon opens and closes the note; a note that holds words or files starts open, and the icon stays lit while it is closed. I need more information is a choice against the options: choosing it clears them, and choosing an option or writing your own answer clears it. Something else opens one line for an answer in your own words, with a picture if that is the answer. An Ask policy can pause an ordinary chat tool call for Deny once or Allow once; approving authorizes only that presented action, and the row reads Waiting for you… until you decide. The pending request is saved with that Message; your decision continues the same tool call. Authorized tool calls, including creating a Package, run in ordinary chat. Ponder asks for a Process only when you want separately managed work. A Question appears in the reply and waits for your answer; saving the answer continues the same Problem. If the connection fails before continuation starts, the answer stays queued for retry. If the Message was saved but the response could not start, the thread reports that failure; send a new Message to continue. Tool steps distinguish a rejected action from an unconfirmed result. A rejection can be corrected. An unconfirmed write asks you to verify the saved state before continuing, so a retry cannot silently repeat the change. Pane previews report access or rendering failures directly instead of treating an error page as a finished preview. A Pane preview follows your selected Changes or version and data reading when no target is specified. Naming another authorized Parable previews that target without inheriting the previous Parable’s draft, Report, or Piece. The camera uses captured code and the settings you can read. Live SQL and Plot requests are not executed inside the captured preview, so a Pane that needs those requests can show its data-host error. If a model sends an invalid tool response, Ponder stops that response and keeps completed actions. Send another Message to continue. The failed step’s speculative text is excluded from the next attempt’s context.

​

Move between conversation and work

Ponder opens conversation-first, filling the height of its column immediately. On desktop, a Parable with additional Pages starts with Ponder at half the available width beside the working surface. Use the left/right arrow to switch between full width and the split; that choice is remembered for the Parable. You can also drag the split. Narrow screens keep the stacked layout. Two more controls sit beside those arrows. The arrow to a line hides Ponder’s column, and the Parable takes the width; the column stays open underneath, so a reply keeps arriving. To bring Ponder back, use the panel button in the Parable’s header, on Ponder’s side: beside the Pin while Ponder is on the right, over the start of the breadcrumbs while it is on the left. It takes no room until the pointer is over that header or over Ponder, and stays shown while Ponder is hidden. Hiding is remembered for the Parable. Ponder starts on the right. The curved arrow moves it to the other side of the Parable: it slides under the Parable and settles there, and the arrow then curves the other way to bring it back. The side is kept in this browser for every Parable. At full width, Pages come into the conversation. Tabs show every Page that renders something, in the order of the Pieces tree: beside the input, stacked from its foot, when the chat is wide enough, and above it otherwise. Many tabs scroll, turning over the edges of their strip. An empty Page (one that places no Pane and whose code draws nothing of its own) has no tab, unless it is the Page you are on. The chosen tab renders the live Page on a card above the thread, and the Page settles in once every Pane is on screen; moving to another tab animates in the same frame, and choosing the open tab again folds the Page away. When a Pane opens another Page, or you use browser Back, the card follows. Drag the card’s bottom edge to resize the Page against the thread; the height is remembered for the Parable. While a Page is shown, the thread rolls away at its top and bottom edges. Pin to Ponder, in a Page’s menu in the Pieces view or in the Pin menu at the Parable’s top right, or asked of Ponder, shows a Page to everyone in the Problem: a Page pinned while you watch opens at once. Pins do not choose the tabs. The fully expanded header shows the Parable’s icon. The Pin menu (the thumbtack beside View as, also in Ponder’s header while it is fully expanded) holds both pins. In the Parable’s header the thumbtack takes no room until the pointer is over the header or its menu is open. Each thumbtack in the menu shows its state: upright while pinned, slashed while not. Pin to Ponder pins or unpins the open Page in the open Problem; Home is always in Ponder. Pin to Navigation for Me keeps the Parable in your sidebar or takes it out. Every active Parable starts pinned, and the choice is yours alone, kept in this browser. Right-click a Parable in the sidebar, or use the menu at the end of its row, for the menu the Pieces view gives a Piece: Move up and Move down reorder your sidebar, and Unpin from Navigation for Me takes the Parable out of it. In Admin Ponder, people who manage Parables also get Rename, which edits the row in place, and Archive. An archived Parable you pinned offers Restore and Delete permanently instead. Archive and Delete ask first, as the Parable’s own menu does. The compact views in its header keep the rest of the work close:

• Pieces shows Pages and Packages together, including Home and each Package’s Panes and Plots. Search covers the complete hierarchy. The Add Package icon sits beside the search field and creates a root Package. Existing Page and Package links continue to open this unified view.
• Adding a Page, Package, Pane, or Plot opens a dialog that asks for its Name and a Description: one sentence on what the Piece is for. A Package or Pane also asks for an icon. The description shows in a band under the selected Piece’s header, where you can change it. Every Page, Package, Pane, and Plot needs one: a save that leaves one without a description is refused (PA-AU-046), and the message names each Piece.
• Collaboration contains Deployments, Participants, and Publishing. Deployments searches and lists Workspace targets; Participants uses the RACI actor picker and assignment rows; Publishing keeps publish actions and patch history.

Problems, Proofs, and Pings remain reachable through existing links but are temporarily absent from the header while those surfaces are unfinished. Select a view to reveal its tree or publishing details above the conversation. Select the Ponder header or diagonal arrow to expand fully up and right; select it again to restore the split width and conversation height. The vertical and horizontal arrows control each direction independently. All three arrows point toward their next action, and behave the same in Pieces and every Collaboration section. Your Problem remains the thread across both layouts. The first breadcrumb uses the Parable’s glyph and Editorial title; later Pieces use the small, muted uppercase Fraktion labels and chevrons of a Custom Parable’s breadcrumb bar. Their glyphs appear when you point at or focus the row. On Admin, the Ponder header holds Search and the Workspace, shown by its mark and name, to the left of Pieces and Collaboration, in every mode. On a Workspace’s own Ponder, the Workspace shows only while Publishing a Parable that is not Personal; in Pinging, Peeking and Personal the header holds Search alone. Dogfood uses the Parable mark, and each Workspace in its menu shows its own. Search opens a cascader across every available context choice and Refresh action: a Workspace, Version, author’s Changes, Refresh, Provider, or action. Choosing Changes also selects their Version. Searching alone changes nothing.

​

Release and Refresh

The top right of the Parable editor reads like a sentence: RELEASE Newest | REFRESH Latest (3d ago), with View as beside it, as its icon, when your seat may use it. While full-column Ponder hides the editor, the same control sits in Ponder’s header. On a phone both captions stay, so Newest and Latest are never confused, and Latest’s age drops. The trail beside it names the Parable and its Pieces only, on both the admin and the Workspace route. Release picks the published Version on screen. It lists the Releases published in this Workspace, on both the admin and the Workspace route. The newest is Newest, the default; older ones go by their numbers, and a line with nothing published is v0. A Workspace that receives the Parable from another Workspace lists only the Releases it published itself, numbered after the source Release it builds on, such as v1.20+1. A newer Release from the source is not in its menu until the Workspace Pulls it and Publishes (Publish & Deploy). Each set of Changes sits indented under the Release it was made from, with its state and age, such as My Changes · Pocketed · 2 days ago, and screen readers hear it inside a group named for that Release. Two sets that would read alike add the clock, or the date, that tells them apart. Hovering a Release shows when it was published, and hovering a set shows when it last changed, in your time zone and clock. The closed control names a set by itself, My Changes, when it was made from the default Release, and as v1.12 · My Changes otherwise. Every mode shows Releases. Every participant sees every set that is not pocketed. A seat that may work in Publishing also sees its own pocketed sets and can start its own; choosing a set moves it to Publishing. A seat that cannot opens a set read-only. The menu has a maximum height and scrolls, and a choice shows on the control at once while the page loads it. Refresh picks the stored Report run, called a Refresh, that the Pages read. Latest is the default: the newest Refresh, with its age, and Pages read the server’s Current for it. When a Pipeline changed and no Refresh has completed since, Latest still reads Current: live Plots read live data, and a stored Plot with no Refresh that fits waits and reads again when one completes. While that Refresh runs, Latest says Refreshing…. For a Parable with no Pipeline, Latest is live Provider data, with no age. When a newer Refresh completes, Latest moves to it and the Pages read again. The menu leads with its actions: Refresh runs the Report now, Preview runs it against your Changes without storing anything, Plan Refreshes schedules it, and Cancel Refresh stops one that is still running. They show only to a seat that may work in Publishing, and choosing one from Peeking or Pinging moves you to Publishing first. During View as they are hidden: the preview shows what the person viewed may do, and nothing on the page runs while it lasts. A refusal for the moment, such as a Refresh already running, keeps the actions visible and says why once below them. Below the actions, Latest, then this month’s Refreshes one line each, such as 3 mins ago or 2 hrs ago with a Failed badge, Older Refreshes in month folders, and a folder for each other Version whose Refreshes the Version on screen cannot read, such as v1.97 · 6 Refreshes: choosing one switches to that Version with it. Two Refreshes that would read alike add the clock, or the date, that tells them apart. Each row carries its exact time on hover and in its spoken name, such as Wednesday, October 7, 2026 at 12:31pm ET. The trigger marks only the Refresh being read: a chosen Refresh names its own state after its age, and Latest is marked when the Refresh it stands for did not succeed. Each row opens the Providers it read. The menu has no search field to read past; arrow keys and typing still find a Refresh. It reads the Refreshes again each time it opens. While a Refresh runs, a slim bar after the group estimates how far along it is, with Est. Time Remaining on hover and Cancel beside it, which shows Cancelling… until the run stops and Refresh cancelled for a few seconds after; a chosen Refresh that failed shows what failed and the way back where the Page was. A choice shows on the control at once. The Pages then read it without holding up the press, and the address records the choice once they show it. The address carries both choices, so a copied link reopens them: version pins a Release and refresh pins a Refresh by its run id. With neither, the page shows the newest Release published in this Workspace, and the Latest Refresh. A refresh the Parable has no Refresh for, such as one that left retention, another Parable’s or a typed word, is dropped once the Refreshes are read, and the page reads Latest. A pinned Failed or Cancelled Refresh stays pinned, and its Pages say it has no rows to read. Moving to another Parable or Workspace drops refresh. A Personal Parable has no Releases, and a Parable that takes no part in reporting has no Refresh control. When nothing has been published yet, Peeking and Pinging have no Version to read, so they open pending Changes instead of an empty page: your own open Changes first, otherwise a colleague’s that your seat on the Parable may open. Any participant opens any set that is not pocketed; a pocketed set opens only for its author. Opening a set never shows you more than the Parable’s current Policies allow: a set that removes or loosens a filter Policy still reads filtered until it is published. With none to open, the page says the Parable is not published yet. Changes are read-only there; switch to Publishing to edit them. Use Participation (the Active Participation Mode), first in the row above the input, when the Parable surface should match what you are doing now. Publishing keeps the authoring surface available, Pinging emphasizes clarification and collaboration, and Peeking keeps a reading-focused view. A mode can hide controls but cannot widen your current Workspace access or Parable responsibility. See Collaboration Modes. Personal Parables save edits directly to main; their dropdown shows Personal and offers Publishing through the saved Collaboration Mode action. The Collaboration panel also offers Enable Publishing & Pinging. The presence roster opens toward the chat content. Plan Refreshes opens the scheduling dialog. Without Ponder access, Search, the Workspace and Participation stay together in one bar above the Pieces; Release and Refresh stay at the editor’s top right.

​

Collaborate without losing the thread

When a connection resumes, Ponder continues after the updates this tab has already received. Delayed live updates do not replay those acknowledged changes. Current access still governs every update; losing access removes the reader from the affected conversation. The presence portraits in the Parable header show who is here. Hover or focus a portrait or the overflow count to reveal one roster of names, recent activity, and locations. The roster stays separate from the neighboring action controls. Everyone in a Problem shares one conversation with Ponder, including what it read. When Ponder runs a query, reads a Piece, or draws a Pane on your turn, that result stays in the Problem’s context, and Ponder can repeat it on anyone else’s turn. Ponder still acts as the person whose Message it is answering, so an action is checked against that person’s access. Keep work that only you should see in a Problem only you can see. When a conversation grows long, Ponder condenses the older part of it and shows a Compacting the conversation… step in the reply. It keeps a summary of what came before and the newest Messages in full. Pictures work the same way: Ponder keeps the most recent ones in view and remembers that older ones were shared. Attach a picture again if Ponder needs to look at it once more. Messages remain in the Problem after you navigate away. Copy a link when a teammate should review a particular message. Use a Ping when you need a person to answer a question or be informed, and use Proofs when the result must be supported by evidence rather than agreement alone. Each of those is a Workspace API operation you can call yourself: a send is Send a Message to Ponder, ending a reply on purpose is Stop a Ponder reply, and a Ping answer is Answer a Problem Ping. The send streams the reply back as it is written, so it is reached through the app’s own server rather than the generated SDKs. Long Problems open at their newest Messages so the conversation is ready quickly. Choose Load earlier Messages at the top of the thread when you need older history; loading a page does not change or summarize the durable Messages. Ponder may propose changes or ask to use an action. Review the target and the effect before approving. A successful response is not the same as a published change: use Publish to inspect audience, responsibility, and readiness.

​

What changes and what stays with the Problem

​

When something does not work

​

Next

Appearance changes synchronize between open app tabs so the theme icon and the next toggle agree with the rendered colors.