Conversations

Every prompt run is recorded, turn by turn, and filed under the feature its terminal was working on. You can read the runs back on a single feature or across the whole project.

What is recorded

A conversation belongs to one terminal. Each turn in it holds the prompt you sent, the files that went with it, the model's reasoning, every tool call with its result and the answer. It also holds what the turn took, which is the model, how long it ran, the tokens in and out, and how many subagents it started.

A terminal starts a new conversation when its context grows too large or when you switch it to another provider. Auto conversations covers when that happens.

Filing

A run is filed under the terminal's binding, which is the WorkMap feature the terminal is on and, when there is one, the requirement it is working on. A run on a terminal that is not bound to anything is still recorded. It has no place on the map.

On a feature

In the WorkMap window, click a card to open its panel. The first tab is Conversations, with a count. Each row shows the provider that did the work, such as Claude Code or Codex, the opening prompt, the date and the number of turns. Click a row to read the whole conversation, and Conversations at the top takes you back to the list.

With nothing recorded the tab reads no conversations recorded for this feature yet.

Across the project

The conversations button on the WorkMap toolbar opens a panel down the right side of the diagram. It is headed Conversations, shows the project's tokens in and out, and lists every conversation recorded for the project, with the most recently active first.

A conversation filed under a feature starts with that feature's position. One that belongs to no feature, such as a question or a run that changed nothing, has no position. Rows open the same way they do on a feature. With nothing recorded the panel reads no conversations recorded in this project yet.

This panel and a feature's panel replace each other, so only one is open at a time.

Reading a transcript

A conversation looks the same whichever provider produced it. Each turn shows these, in order.

• The day and time the prompt was sent.
• The prompt in a box, with a button that keeps it as a saved prompt and a chip for each attached file, which opens the file.
• The run itself, with reasoning under a Thinking label, each tool call with its output, and the model's own text.
• A line with the model, how long the turn took, the tokens in and out, the number of subagents, and a note when the turn was stopped.
• A chip for each skill the turn used.

A prompt the app sent by itself, such as an automatic continue after a provider error, shows as Auto continued rather than as something you typed. See Auto continue.

Where they live

Conversations are kept in a database on your own machine, inside the data folder. Prompts, the run's content and the provider's raw output never leave that database. See The data folder.