v0.1.0 — first public release

promptkit v0.1.0

The first public release of promptkit — a TypeScript-first toolkit for defining, rendering, testing, and evaluating LLM prompts.

Highlights

• definePrompt — typed, validated, versioned prompts.
• runPrompt — provider-agnostic execution. You bring the model; promptkit never calls one itself.
• evalPrompt — run a prompt against a suite of cases with assertions.
• expect — a small assertion library tuned for LLM output (toContain, toMatch, toMatchJSON, toBeOneOf, …).
• renderTemplate — mustache-style interpolation with dot paths.
• loadPromptsFromDir — discover prompts from a directory.
• promptkit CLI — list, run, eval, version, help.

Philosophy

• Zero runtime dependencies.
• Provider-agnostic (OpenAI, Anthropic, local models, mocks — all the same to promptkit).
• Small, composable core.
• MIT licensed, community-maintained, non-commercial mission.

Install

npm install promptkit

What's next

See the open issues for roadmap items — including schema adapters (Zod/Valibot), parallel evals, richer reporters, token/cost estimation, and a portable prompt-pack format. Good first issues are labeled; PRs welcome!

Full changelog: https://github.com/Cryptoteep/promptkit/blob/main/CHANGELOG.md