Open-source infrastructure for continuous learning
Work in the open-source harness or web app. Turn useful outcomes into Tasksets, evals, and training runs, then improve your model and repeat.
OpenPond keeps the task, execution environment, evidence, review, and reward connected. Turn useful outcomes into Tasksets, evals, and training runs without reducing the work to a pile of chat logs.
Work in Desktop, CLI, TUI, or the web app beside the files, terminal, browser, review, and teammates the job needs. Stay local or move execution to an isolated cloud sandbox without changing the underlying work.
OpenPond turns selected work into an inspectable loop. Keep the grader, trace, baseline, reward, model version, and deployment decision together—so “better” means more than a nicer demo.
Use Work Mode with local projects or cloud sandboxes. The conversation, tools, files, diffs, and outcome stay together.
Turn useful runs into Tasksets with explicit inputs, outcomes, splits, verifiers, and graders you can inspect.
Run evals, revise the agent or harness, and train a task-specific model only when the evidence says it is worth it.
Managed reinforcement learning
Compare a baseline, inspect reward variation, run bounded GRPO or RFT, and preserve the resulting adapter with its complete recipe. A failed hypothesis is still a useful result.
Explore Tasksets →OpenPond Desktop, CLI, TUI, local runtime, Tasksets, and the training workbench live together. Run locally without an OpenPond account, inspect the contracts, and bring in the web app when the work needs cloud execution or a team.
Read the documentation →Start from the terminal
npx openpond@latestThe same source-backed agent can run beside a local project, continue in OpenPond Cloud, or meet teammates in the channels they already use.