# Exporting Chat Logs from Common LLM Logging Systems

Canonical: https://hyperparam.app/docs/exporting-chat-logs

This guide summarizes how to pull your agent traces, coding-tool transcripts, and chatbot histories out of popular logging and observability platforms so you can bring them into Hyperparam. Once exported, drop them in directly or stage them in S3, GCS, or Azure to stream from the browser. Hyperparam can join logs from multiple platforms in a single workspace.

> If you are setting up AI log collection from scratch rather than migrating off an existing vendor, see the architecture overview for how the pieces fit: HypAware, the improvement loop for all your agents, records sessions as Apache Iceberg tables, and Hyperparam analyzes them.

## Claude Code

- Claude Code writes conversation transcripts to `~/.claude/projects/<project-hash>/<session-id>.jsonl`.
- These JSONL files contain full conversations including messages, responses, and tool calls.
- See [Claude Code OpenTelemetry Setup](https://hyperparam.app/docs/claude-code-otel) to export traces via OTel.
- See [Claude Code Logging with LiteLLM](https://hyperparam.app/docs/claude-code-litellm) to log via a proxy.

## ChatGPT

- Go to **Settings → Data controls → Export data** and click **Export**.
- OpenAI will email you a link to download your archive (~5 minutes).
- The archive contains a `conversations.json` file with your full chat history.

## Langfuse

- Use the **Exports** tab in the Langfuse UI to download data as CSV.
- Available tables: `traces`, `datasets`, and `audit logs`.
- Select the dataset, choose the file format, and click **Export** to start a one-time download.

## Arize Phoenix

- Navigate to the tracing interface, then open the **Export** panel.
- Phoenix supports CSV downloads for trace tables.

## LangSmith

- Set up **Bulk Data Export** under **Settings → Integrations**.
- Configure a destination such as BigQuery, Snowflake, AWS S3, or other supported warehouses.
- The service pushes trace data on a schedule.

## Datadog

- Enable **Log Archives** for your Datadog organization and choose S3 or GCS as the sink.
- Archives arrive as `.json.gz` files partitioned by time.
- For ad hoc pulls, open **Logs Explorer**, run a query, and click **Export** to retrieve a JSON snapshot.

## Grafana Loki

- Install or open `logcli`, Grafana Loki's command-line client.
- Run queries locally and redirect the response to disk:

  ```bash
  logcli query '{app="llm"}' --limit=5000 --output=json > llm-traces.json
  ```

- Use `--output=raw` if you prefer plain log lines.

## Honeycomb

- After running a query in the Honeycomb UI, choose **Actions → Download Results** to export CSV.
- For large result sets, paginate through time windows or use the API to fetch segments iteratively.
- The [Events API](https://docs.honeycomb.io/api/events/) supports programmatic extraction.