Sessions and the transcript

In DataMind Oda, a chat is one session: the unit of work that holds everything done in it. This page explains what a session records, the parts of the chat header, how a turn runs and ends, and what a branched chat carries over.

A chat is a session

• Every chat you open is one session. As you work, Oda adds to it in order: your messages, its answers, every step it takes and every result the step produces.
• The session only grows. What has been saved is never rewritten: Oda adds new entries instead of changing old ones.
• The transcript you read is built from this record. Oda also uses the word session in places such as Download session log, which saves the chat's record as a file (Download a chat log).

The chat header

The top of a chat holds everything about the chat itself:

• Session hierarchy: the chain of chats this one belongs to, ending with the chat you are reading. The helper chats Oda started for this one appear in the bar too, and you can switch between them there.
• The chat's title.
• The mode the chat runs under, shown as plain text, for example Minimal mode. A chat keeps the mode it started with.
• A count such as 1 subagent shows how many helper chats Oda has started for this chat.
• More actions (…), the chat's menu: whole-chat rows live there, such as downloading the chat's log.
• Open right sidebar brings out the side panel beside the conversation (Chat surfaces and status).

How a turn flows

A turn starts when Oda takes one of your messages and begins working on it:

1. Oda plans and then works in steps. A step is one pass of thinking plus the actions it asked for, and every action appears in the chat as a step row (Oda's steps and tool results).
2. The answer arrives as Oda writes it, so a long reply fills in while the turn is still running.
3. If a step needs your permission, an approval card appears in the conversation and you answer it in place; Oda carries on with your answer (Approvals).
4. When nothing is left running, the turn settles and the message box is ready for your next message. A chat keeps one story through all of its turns.

How a turn ends

A turn can end in four ways:

• It finishes. The answer stands, and the chat is ready to carry on.
• You stop it. The reply keeps the part that had already appeared, and the turn is marked Stopped; Oda does not continue it (Stopping Oda and queued messages).
• It fails. The turn reads This turn failed, with the reason named beside it. Try the request again, and see Troubleshooting if it repeats.
• It reaches the output limit. The reply stops with Output token limit reached: The reply was cut off; earlier output is preserved in the conversation. Send "continue" to let the model resume.

The transcript

The conversation is the session's transcript. It carries the whole story in order: your messages, Oda's answers, a step row for every action, and the cards that gather what a turn produced.

• Oda's answers are written to be read: headings, lists, tables, code, formulas, images and links all render in place (Reading Oda's answers).
• Under a finished answer are the message actions: Copy takes the answer's text, and Good response and Bad response record what you thought of it, where reply ratings are shown.
• The rail beside the conversation outlines its turns, and the statistics strip counts them (Chat navigation and stats). The Trajectory view replays the same session record with its timings (Trajectory and session logs).

Branches and inherited history

• On the last message of a completed turn you find Branch into a new conversation: it starts a separate chat from that message, with the conversation up to that point already in it as inherited history.
• The chat you branched from is not changed: its history stays as it was, and later work in the branch does not appear in it.
• A branch cut while a turn was still running starts without that turn's unfinished results.
• Editing, rewinding or retrying inside one chat is a different move: it keeps one chat and adds a version to it (Conversation branches).

GitHub review chats

A chat opened for a GitHub review is easy to spot:

• It is titled Review <repo>#<number>, naming the repository and the number under review.
• It carries a GitHub event received badge and opens in Read Only.
• When the review is done, the chat ends with a short result, such as Review complete: no actionable findings.

How a GitHub delivery starts a chat is covered in Webhook sessions.

When a chat grows long

• When a long chat fills the model's reading space, Oda shortens the older part automatically and carries on from the summary; you can also run /compact yourself (Chat with Oda).
• Shortening changes what the model reads, not what the record keeps: the earlier entries are hidden from the model, not deleted, and nothing that has been saved is ever rewritten.

Questions, answered

Is a session the same as a chat?

Yes. A chat is one session: the unit of work that holds everything done in it. Oda uses the word session in places such as Download session log.

What does the Session hierarchy bar show?

The chain of chats the open chat belongs to, ending with the chat you are reading. The helper chats Oda started for this one appear in the bar too, and you can switch between them there.

What is the difference between a turn and a step?

A turn is one round of work, started by your message; a step is one part of it, made of a single pass of thinking plus the actions it asked for. A turn can hold several steps, and each step shows as its own row in the chat.

How do I know when a turn has finished?

The work settles: the reply stops growing, the step rows stop running, and the message box takes your next message.

Does branching a chat change the chat it came from?

No. The new chat carries the conversation up to that message and continues on its own; the chat you branched from keeps its history, and later work in one does not appear in the other.

What is a GitHub review chat?

A chat opened for a GitHub review. Its title reads Review <repo>#<number>, it carries a GitHub event received badge, it opens in Read Only, and it ends with a short result when the review is done.

Does shortening a chat delete the earlier messages?

No. Shortening changes what the model reads, not what the record keeps: the earlier entries are hidden, not deleted, and nothing that has been saved is rewritten.

Where can I see everything that happened in a chat?

The transcript keeps it in order: your messages, Oda's answers and a row for every step, which you can expand. For the record with its timings, open the Trajectory view (Trajectory and session logs).