# OpenPond

Open-source workspace where people and agents complete real work. OpenPond gives agents access to the project files, tools, terminal, browser, and cloud sandboxes needed to finish a task. The conversation, tools, files, changes, and result stay with the project.

## When to use OpenPond

- Complete a software, research, operations, or content task that needs project context and tools.
- Work beside local files, or use an isolated cloud sandbox when the task needs more room.
- Review completed work and turn evidence-backed improvements into testable Harness updates.
- Run the same source-backed agent across Desktop, web, CLI, TUI, team chat, Slack, or Microsoft Teams.

## Start here

- [OpenPond Cloud](https://openpond.ai/login): Start or continue a task in the hosted workspace.
- [OpenPond documentation](https://openpond.ai/docs): Product and developer documentation.
- [OpenPond source repository](https://github.com/openpond/openpond): Desktop, CLI, TUI, local runtime, Tasksets, and training workbench.
- [OpenPond CLI](https://openpond.ai/products/cli): Run `npx openpond@latest` to start from the terminal.

## API and authentication

- [OpenPond public API](https://api.openpond.ai): API-key and account-authenticated endpoints for OpenPond products, sandboxes, and OpChat.
- [API health](https://api.openpond.ai/health): Unauthenticated service health status.
- Use an OpenPond account or API key for protected API operations. Do not treat public documentation as authorization to access private projects, accounts, prompts, or data.

## Documentation

- [Overview](/docs/workspace): Navigate the Console workspace, evaluation, reinforcement learning, and services.
- [Profile](/docs/workspace/profile): Manage reusable agent source, skills, workflows, and evaluations.
- [Marketplace](/docs/workspace/marketplace): Browse datasets, request custom data, and review your marketplace portfolio.
- [Connected agents](/docs/workspace/connected-agents): Connect agent activity and explicitly import completed conversation evidence.
- [Inbox](/docs/workspace/inbox): Review pending human decisions in the current workspace.
- [Choose a surface](/docs/workspace/choose-a-surface): Choose Console, Chat, Work, an API, or the local workbench.
- [Chat and Work](/docs/workspace/chat-vs-work): Choose conversational help or a reviewable multi-step work result.
- [Skills](/docs/workspace/skills): Package a focused workflow in SKILL.md and invoke it intentionally.
- [Agent SDK](/docs/workspace/agent-sdk): Build, validate, evaluate, and package TypeScript-first OpenPond Agents.
- [Projects](/docs/workspace/work-projects): Create a durable Cloud project from a repository or an empty starting point.
- [Evaluation overview](/docs/evaluation): Measure models and agents with datasets, graders, and experiments.
- [Datasets](/docs/evaluation/datasets): Author, import, review, and publish tasks for experiments and training.
- [Graders](/docs/evaluation/graders): Define repeatable criteria for measuring task outcomes.
- [Experiments](/docs/evaluation/experiments): Run fresh attempts or score recorded evidence against pinned dataset and grader releases.
- [Review and label tasks](/docs/evaluation/human-review): Apply a grader's criteria to responses and decide which tasks are useful for learning.
- [Improve an LLM judge](/docs/evaluation/llm-judges): Define a rubric, add golden examples, and check the judge against human labels.
- [Use a code verifier](/docs/evaluation/code-verifiers): Score checkable outcomes with rules, tests and expected state.
- [Change a grader safely](/docs/evaluation/grader-versions): Improve scoring between learning runs while preserving fair comparisons and history.
- [Environments](/docs/evaluation/environments): Inspect the versioned runtime requirements attached to published datasets.
- [Training](/docs/reinforcement-learning): Evaluate tasks, review results, improve a grader, and train a candidate model.
- [Continual Learning](/docs/reinforcement-learning/continual-learning): Import local agent conversations, choose graders, and schedule training with independent holdout evaluations.
- [Models](/docs/reinforcement-learning/models): Inspect trained versions, candidate lineage, evaluations, and deployment decisions.
- [Train a model in OpenPond Desktop](/docs/reinforcement-learning/desktop-training): Turn reviewed chats into a Taskset, train a model, evaluate it, and decide whether to use it.
- [Services overview](/docs/services): Find the Console services for inference, activity, compute, automation, and administration.
- [LLM Gateway](/docs/services/llm-gateway): Call OpenPond hosted models through one OpenAI-compatible gateway.
- [OpenAI-compatible API](/docs/services/llm-gateway/openai-compatible): Configure an OpenAI client for OpenPond hosted model inference.
- [OpenAI JavaScript quickstart](/docs/services/llm-gateway/openai-javascript): List hosted models and make a chat completion with an OpenPond API key.
- [Hosted model catalog](/docs/services/llm-gateway/catalog): Find public hosted model IDs, lifecycle, context windows, capabilities, and token rates.
- [OpenAI Python quickstart](/docs/services/llm-gateway/openai-python): Make a hosted OpenPond model request with the OpenAI Python SDK.
- [cURL quickstart](/docs/services/llm-gateway/curl): List hosted OpenPond models and create one completion over HTTP.
- [Stream a model response](/docs/services/llm-gateway/streaming): Consume OpenAI-compatible streaming output from a hosted OpenPond model.
- [Tool calling](/docs/services/llm-gateway/tool-calling): Send tool declarations and handle returned tool calls with an OpenAI-compatible model request.
- [Ponder](/docs/services/ponder): Inspect private recorded activity and select evidence for evaluation or learning.
- [Sandboxes](/docs/services/sandbox-compute): Run agentic work or control an isolated OpenPond sandbox from server-side TypeScript.
- [Work, sandboxes, and runtimes](/docs/services/sandbox-compute/work-sandboxes-and-runtimes): Choose between model-driven Work, direct sandbox control, and durable workflow runtimes.
- [Files, processes, and previews](/docs/services/sandbox-compute/files-processes-and-previews): Write files, run commands and services, and expose bounded previews with the Sandbox SDK.
- [Pricing and budgets](/docs/services/sandbox-compute/pricing-and-budgets): Understand per-second sandbox pricing, resource tiers, included usage, and hard spend controls.
- [Workflow modes](/docs/services/sandbox-compute/workflow-modes): Select the runtime policy for attempts, features, scheduled work, replay, evaluation, and review.
- [Runtimes, snapshots, and cleanup](/docs/services/sandbox-compute/runtimes-snapshots-and-cleanup): Build durable sandbox workflows with events, checkpoints, snapshots, and explicit retention.
- [Serving](/docs/services/serving): Review personalized model usage, deployments, and API access.
- [Persistent Volumes](/docs/services/persistent-volumes): Manage retained workspace files separately from sandbox execution.
- [App Hosting](/docs/services/app-hosting): Inspect published applications, source projects, builds, and releases.
- [Databases and Infrastructure](/docs/services/infrastructure): Review workspace database configuration and supported cloud installation controls.
- [Harness](/docs/services/harness): See how OpenPond reviews completed work, proposes bounded improvements, and keeps every release inspectable and reversible.
- [Continuous learning](/docs/services/harness/continuous-learning): Review recurring work, improve the Harness, and decide whether controlled Evaluation or model training is warranted.
- [Refiner evidence](/docs/services/harness/refiner-evidence): See the bounded public evidence packet used for a completed-turn refinement review.
- [Refiner decisions and routing](/docs/services/harness/refiner-decision-and-routing): Understand model-owned relevance, no-action outcomes, and smallest-layer routing.
- [Refiner critique and validation](/docs/services/harness/refiner-critique-and-validation): Learn how proposals are generalized, checked, and rejected before application.
- [Refiner application and rollback](/docs/services/harness/refiner-apply-and-rollback): Follow a validated proposal into an immutable release transition and back out safely.
- [Refiner receipts](/docs/services/harness/refiner-receipts): Read observable refinement activity without exposing private reasoning or private evidence.
- [Harness privacy and authority](/docs/services/harness/privacy-and-authority): Understand evidence scope, redaction, approval gates, and what never becomes Harness source.
- [Workflows](/docs/services/workflows): Run a saved prompt or immutable Agent action with clear triggers and authority.
- [Schedules and webhooks](/docs/services/schedules-and-webhooks): Trigger a Workflow with a bounded contract and explicit authority.
- [Integrations](/docs/services/integrations): Connect provider accounts and manage OAuth and MCP access for your workspace.
- [Identity & Access](/docs/services/identity-access): Manage workspace membership, roles, API keys, and settings.
- [Secrets Manager](/docs/services/secrets-manager): Manage workspace environment configuration and the lifecycle of runtime secrets.
- [Billing](/docs/services/billing): Inspect workspace billing, service usage, and run spending limits.
- [Logs](/docs/services/logs): Follow resource-specific execution, build, release, and request evidence.
- [Desktop & Open Source](/docs/services/open-source): Run OpenPond locally, use terminal interfaces, and inspect the public Harness.
- [Use OpenPond Desktop](/docs/services/open-source/desktop): Keep local projects, reviewable work, and improvement evidence in one workbench.
- [Open-source Harness](/docs/services/open-source/harness): Understand the public provider-neutral package behind releases, refinement, receipts, and rollback.
- [CLI and TUI](/docs/services/open-source/cli-and-tui): Use OpenPond's open-source runtime from a terminal interface.
- [Contribute to OpenPond](/docs/services/open-source/contributing): Propose a safe, reviewable change to public OpenPond source.
- [Authentication and scopes](/docs/services/reference/authentication-and-scopes): Map OpenPond API requests to the smallest required API-key scope.
- [Create an API key](/docs/services/reference/api-keys): Create, scope, store, rotate, and revoke an OpenPond API key.
- [Models API reference](/docs/services/reference/models-api): Find hosted OpenAI-compatible model endpoints.
- [Sandbox API reference](/docs/services/reference/sandbox-api): Use the server-side SDK as the preferred contract for isolated compute.
- [Errors and limits](/docs/services/reference/errors-and-limits): Handle public API failures, retries, and resource limits safely.
- [Availability labels](/docs/services/reference/availability): Read stable, preview, enterprise, and unavailable feature labels consistently.