Chat logging

DataMind Oda keeps each chat as a log file on your computer, and the same log travels to your company's DataMind server with each request the chat makes, where it is recorded in the company's Observability log. This page covers what travels with a request, how delivery works, and how logging can be turned off. The file side is on Session logs on your computer; for a copy you keep yourself, see Download a chat log.

What travels with a request

Every request a chat makes carries the chat's own log:

• The chat's record. Your messages and Oda's answers, what Oda's steps ran and returned, any feedback you leave, and the instructions and settings the chat works under, plus the folder it works in.
• The record travels whole. The log's own entries travel as they were recorded, not a summary; nothing is filtered out of the copy before it travels.
• Credentials are not part of it. Oda's own stored model credentials do not travel with the log. Anything that appears in the chat itself, such as a password typed into a message, is part of it.
• The header. The opening part of the log that names the chat and how it runs; Chat log format lists its fields.
• No channel of its own. The log does not leave by a separate route: it rides inside the requests the chat already makes, and it goes only to the server you signed in to.

The company records it in its Observability log; Manage DataMind Oda for your company describes that log from the company side. The uploads that feedback releases are a separate channel; Feedback and privacy covers them.

How delivery works

• Incremental. The first request carries the chat's log from its start; after that, each request carries only the events newer than the last point the server accepted.
• With the request, before the answer. The log is part of the request itself, so the server takes it before the answer starts to stream.
• The point moves only after the server accepts. Oda continues from a later point only once the server has accepted what it sent, and the point never moves backwards.
• Repeated, never skipped. If a request fails, the point does not move, and the same part is carried again by a later request. If a sudden stop leaves it unclear whether a part arrived, that part can be sent twice.
• No separate size limit. The log part has no size limit of its own: a request carries the new part whole, and a request the server refuses is retried later rather than shortened.
• Each chat keeps its own place. A branch does not take on the place its parent chat had reached; each chat tracks its own, so a branch's log is carried from its start.

Turning logging off

Logging is on by default: a chat's requests carry the log unless the installation is set up to stop it. It can be turned off for one person or for a whole installation.

• For yourself. If you have the Disable DataMind Oda Content Logging permission, turn content logging off in your DataMind OS profile. The company log then keeps only who made each request, when, with which model and how many tokens it used. Manage DataMind Oda for your company covers the permission.
• For a whole installation. Oda can also be set up so that its requests carry no log copy at all. That is an installation setting; ask your administrator if you need it.

Questions, answered

What of my chat is sent to the company?

The chat's own log, whole and unfiltered: your messages and Oda's answers, what Oda's steps ran and returned, any feedback you leave, the instructions and settings the chat works under, and the header that names the chat. It goes with each request the chat makes, and the company records it in its Observability log. Chat log format covers the log's contents in full.

Is my whole chat sent with every request?

No. The first request carries the log from its start; after that, each request carries only the events newer than the last point the server accepted.

When is the log sent?

It travels inside the request itself, so the server takes it before the answer starts to stream.

What happens if a request fails?

The point Oda continues from does not move, so the same part of the log travels again with a later request. A failed request never causes a gap.

Can the same part of a chat be sent twice?

Yes, in one case: if a sudden stop leaves it unclear whether a part reached the server, Oda sends that part again rather than risk a gap. Delivery is built to repeat, never to skip.

Is there a size limit on the log part of a request?

Not of its own: a request carries the new part whole. If the server refuses the request, the part is retried later, and the log is never shortened to make a request fit.

Does a branch continue from where its parent chat left off?

No. Each chat tracks its own place. A branch does not take on the place its parent chat had reached, so it does not skip ahead based on what was already sent from that chat.

Does the log include my model credentials?

No. Oda's own stored model credentials are not part of a chat's log. Anything that appears in the chat itself, such as a password typed into a message, is part of it and travels with the log.

How do I stop my requests from being logged?

If you have the Disable DataMind Oda Content Logging permission, turn content logging off in your DataMind OS profile. The company log then keeps only who made each request, when, with which model and how many tokens it used. Your requests still reach the server so they can be answered; only what is recorded changes.

Can I turn logging off for just one chat?

No. Content logging is a personal setting, not a per-chat one: it covers your requests as a whole.

Can the log be turned off for everyone?

Yes. Oda can be set up so that its requests carry no log copy at all. That is an installation setting; ask your administrator.