DataCLI turns structured prompt records into validated chat datasets or agent-harness traces. The project is split into a CLI, an SDK, and a framework-independent runtime.
- Getting started covers installation, prompt creation, generation, and output inspection.
- CLI reference documents every command, option, default, and exit code.
- Configuration files explains YAML structure, command sections, validation, path resolution, and CLI precedence.
- Prompts and output format defines the input JSONL schema, multi-turn prompts, message types, metadata fields, ordering, and partial files.
- Architecture and execution model explains packages, jobs, runs, concurrency, cancellation, validation, and disposal.
- Environments and agent traces covers current-directory and isolated filesystem workspaces, retention, seed directories, and Pi traces.
- SDK and custom pipelines covers adapters, chat pipelines, tools, hooks, programmatic jobs, and local pipeline modules.
- Providers documents OpenRouter and generic OpenAI-compatible endpoints, request parameters, timeouts, response normalization, pricing, and ephemeral keys.
- Operations and troubleshooting covers safe reruns, output conflicts, failures, signals, error codes, credential handling, and common problems.
- Development describes the monorepo, commands, TypeScript build graph, tests, and package publishing checks.
| Goal | Recommended entry point |
|---|---|
| Generate ordinary assistant conversations | datacli generate |
| Capture coding-agent activity and tool use | datacli trace |
| Run a project-specific pipeline module | datacli run |
| Embed dataset generation in TypeScript | @teichai/datacli-sdk |
| Build directly on runtime contracts | @teichai/datacli-core |