Import agent activity

The importer turns selected agent conversations into saved evidence you can review for Datasets and training. It supports an explicit one-time upload or recurring imports; connecting a source alone does not upload its history.

Choose a source

Open Connected agents to configure where conversation evidence comes from. Supported source cards include Pal, Codex, Claude Code, OpenCode, Grok Build, Hermes, Pi, Oh My Pi, and OpenClaw. The OpenPond card switches between Web (Pal) and Local (OpenPond local). A custom Agent import is also available.

Connect a source

1. Choose the source card and open its setup.
2. Follow its source-specific collection or export instructions on the machine that owns the conversation.
3. Choose the workspace and destination offered by the setup, and review the collection scope before enabling access.
4. Inspect status and imported tasks before using them in an experiment.

The source setup provides an official CLI command with the destination selected. Connecting authorizes that destination without uploading history. Choose specific conversations to upload, or explicitly request bulk history with --upload. Continual imports default off; --continual enables a separate import schedule. The continual learning setup guide explains source commands, scope, scheduling and status. Importing does not configure model training.

Manual file imports use structured session exports. Preserve related files and explicitly selected branches or child transcripts when required. Finish the turn before using it as completed evidence.

Import once or collect new activity

The connection flow supplies commands with your selected destination. To inspect available operations first, run openpond import --help. Reuse the account and OpenPond home configured in Desktop; pass that home with --home.

For example, connect a Codex source without uploading history:

For an explicitly approved bulk import of the past week and daily future imports:

Review the destination and collection scope during setup. Use the connection status and imported task list to verify receipt. The import schedule collects new activity; it does not enable a Model's training schedule.

Review before import

The import preview shows the selected conversation and missing context. Check that it contains the intended turn, tool evidence, and outcome. Exclude secrets and unrelated work. Importing records evidence; it does not approve that evidence for training or assert that the agent succeeded.

Use Recorded activity to inspect recorded activity and choose cases for a Dataset or Experiment.

Manage access and recovery

Open the source's details or setup controls for collection status and recovery. Use its connection controls to add a different computer or agent profile. Disconnecting revokes future access from that source; saved tasks remain. If an import reports incomplete context, correct the export or narrow the selection before treating it as reproducible evidence.

Use imported evidence for learning

Choose useful cases for a Dataset, review the inputs and outcomes, attach checked graders and keep training examples separate from held-out evaluation tasks. Then set up continual learning to choose a compatible Model, review rules, schedule and budget.

For individual tool failures and recovery evidence, use Capture tool errors with Fix-it.