Skip to content
OpenPond
Join community
Overview/Choose your starting point

Choose your starting point

Choose a model workflow and adapter, continual learning, evaluation or serving, then the tools you need.

Choose the model outcome you need, then open the corresponding workflow in Console. The guides retain the exact data, model and evaluation boundaries behind each action.

Train a model

Use Ponder Training for reviewed SFT demonstrations or reward-scored GRPO tasks, or choose a Decision Model for structured choice training and scoring. Prepare the Dataset, baseline and independent evaluation first. Review runtime readiness and the maximum spend before starting the exact preparation.

Choose an adapter

Use the adapter guide to compare LoRA, Engram and Neuron training beneath the selected workflow. Its support matrix records which pairs are currently executable. All adapter choices remain visible and selectable for authoring; unsupported pairs cannot execute.

Learn as new data arrives

Use Continual learning to configure the Model's source, review rules, schedule, budget and candidate behavior. Start from the Model or a Dataset's Continual learning tab; both use the same policy. A scheduled check waits until enough eligible data is available. For coding-agent evidence, follow Continual learning setup.

Evaluate and review

Use Datasets, Graders and Experiments to establish a baseline and compare candidates under the same rules. Review exact evidence and keep training tasks out of the holdout. Use the Model's Tasks queue for ordinary admission and Inbox for assignments created by the supported workflows.

Serve or call a model

Use Model deployments to review and activate a compatible saved version. Use Inference to call available hosted models through the OpenAI-compatible API. A saved artifact is not itself a callable deployment.

Connect an application or local tools

The SDK integrates workflows in your application. Harness manages agent behavior around a model. Desktop keeps local projects, conversations and reviewable outputs together; coding-agent setup connects an external agent. Use Chat and Work for assistant tasks and Sandboxes for isolated compute.

API integrations start with Create an API key. For complete examples, see Tutorials.

View as Markdown