> ## Documentation Index
> Fetch the complete documentation index at: https://docs.trymonte.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Overview

> Improve a deployed model from real use

A deployed model meets requests that no eval covered. Continual learning turns that traffic into the next round of training. A deployment records each request as a trace. The team reviews the traces, turns the useful ones into dataset rows or tasks, trains from the deployed checkpoint, and ships the result behind the same endpoint.

## In This Section

| Page | Covers |
| - | - |
| [Traces](https://docs.trymonte.ai/continual-learning/traces.md) | The live requests and telemetry a deployment records |
| [Continuous Retraining](https://docs.trymonte.ai/continual-learning/continuous-retraining.md) | Feed reviewed traces into datasets and environments, round by round |

## One Round

1. **Collect.** A deployment keeps each request and its response as a trace. Collection is on by default, and the traces stay in artifact storage, which is in the team's own account with BYOC. `--traces off` on a deployment keeps none.

2. **Review.** Download recent traces and pick the ones worth learning from: wrong answers, hand-offs that a person overturned, and new kinds of requests.

   ```bash theme={null}
   monte trace download --deployment support-triage --since 7d -o week.jsonl
   ```

   A trace records what the model did, not whether it was right. The app can tag each request with a `metadata` field, such as a ticket ID, so the review joins traces with the outcome in the team's own ticketing system.

3. **Turn them into data.** Each trace is a chat `messages` row, the same format as an SFT dataset.

   * A corrected reply becomes a demonstration row in a new dataset version.
   * A reviewer's edit and the model's reply become a [preference pair](https://docs.trymonte.ai/training/dpo.md).
   * A new kind of request becomes a task in the [training environment](https://docs.trymonte.ai/environments/overview.md).

   ```bash theme={null}
   monte dataset add support-prefs --from-file week-pairs.jsonl --kind preference --anonymize
   ```

4. **Retrain.** Start a run from the deployed checkpoint, `train-11/300`, with the same method that made it. The new dataset version holds the earlier rows and the new ones, so the model keeps what it learned.

   ```bash theme={null}
   monte train support-triage --recipe triage-dpo --data support-prefs@2 --steps 100 \
     --parent train-11/300 --remote
   ```

5. **Ship.** Score the new checkpoint on the development eval, and [compare it](https://docs.trymonte.ai/evaluation/metrics.md) task by task with the deployed checkpoint. Ship it only if it fixes more tasks than it breaks, then [move traffic to it](https://docs.trymonte.ai/serving/scaling-versions.md) in steps.

Each step is a CLI command, so a team runs a round on its own schedule, or Monte's engineers run it with the team.

An experiment's evals never change, so new tasks in an eval need a new experiment. Training data can change in every round.

## Privacy

`--anonymize` replaces personal details in the dataset rows before they are stored. The traces themselves keep the original text in artifact storage, and `--traces off` keeps none. [Anonymization](https://docs.trymonte.ai/datasets/anonymization.md) lists what it finds. A team's traces train only that team's models.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.