🚀 JavaScript-ish environment with convenient tooling for file ingestion, prompt development, and structured data extraction.
- 📄 Read the ONLINE DOCUMENTATION at microsoft.github.io/genaiscript
- 🎙️ Listen to the podcast (generated by NotebookLM).
overview.mp4
Programmatically assemble prompts for LLMs using JavaScript. Orchestrate LLMs, tools, and data in a single script.
- JavaScript toolbox to work with prompts
- Abstraction to make it easy and productive
- Seamless Visual Studio Code integration
Say to you want to create an LLM script that generates a 'hello world' poem. You can write the following script:
$`Write a 'hello world' poem.`The $ function is a template tag that creates a prompt. The prompt is then sent to the LLM (you configured), which generates the poem.
Let's make it more interesting by adding files, data and structured output. Say you want to include a file in the prompt, and then save the output in a file. You can write the following script:
// read files
const file = await workspace.readText("data.txt")
// include the file content in the prompt in a context-friendly way
def("DATA", file)
// the task
$`Analyze DATA and extract data in JSON in data.json.`The def function includes the content of the file, and optimizes it if necessary for the target LLM. GenAIScript script also parses the LLM output
and will extract the data.json file automatically.
Get started quickly by installing the Visual Studio Code Extension or using the command line.
Build prompts programmatically using JavaScript or TypeScript.
def("FILE", env.files, { endsWith: ".pdf" })
$`Summarize FILE. Today is ${new Date()}.`Edit, Debug, Run, and Test your scripts in Visual Studio Code or with the command line.
Scripts are files! They can be versioned, shared, and forked.
// define the context
def("FILE", env.files, { endsWith: ".pdf" })
// structure the data
const schema = defSchema("DATA", { type: "array", items: { type: "string" } })
// assign the task
$`Analyze FILE and extract data to JSON using the ${schema} schema.`Define, validate, and repair data using schemas.
const data = defSchema("MY_DATA", { type: "array", items: { ... } })
$`Extract data from files using ${data} schema.`def("PDF", env.files, { endsWith: ".pdf" })
const { pages } = await parsers.PDF(env.files[0])Manipulate tabular data from CSV, XLSX, ...
def("DATA", env.files, { endsWith: ".csv", sliceHead: 100 })
const rows = await parsers.CSV(env.files[0])
defData("ROWS", rows, { sliceHead: 100 })Extract files and diff from the LLM output. Preview changes in Refactoring UI.
$`Save the result in poem.txt.`FILE ./poem.txt
The quick brown fox jumps over the lazy dog.Grep or fuzz search files.
const { files } = await workspace.grep(/[a-z][a-z0-9]+/, { globs: "*.md" })Register JavaScript functions as tools (with fallback for models that don't support tools).
defTool(
"weather",
"query a weather web api",
{ location: "string" },
async (args) =>
await fetch(`https://weather.api.api/?location=${args.location}`)
)Register JavaScript functions as tools and combine tools + prompt into agents.
defAgent(
"git",
"Query a repository using Git to accomplish tasks.",
`Your are a helpful LLM agent that can use the git tools to query the current repository.
Answer the question in QUERY.
- The current repository is the same as github repository.`,
{ model, system: ["system.github_info"], tools: ["git"] }
)then use it as a tool
script({ tools: "agent" })
$`Do a statistical analysis of the last commits`const { files } = await retrieval.vectorSearch("cats", "**/*.md")Run models through GitHub Models or GitHub Copilot.
script({ ..., model: "github:gpt-4o" })Run your scripts with Open Source models, like Phi-3, using Ollama, LocalAI.
script({ ..., model: "ollama:phi3" })Let the LLM run code in a sandboxed execution environment.
script({ tools: ["python_code_interpreter"] })Run code in Docker containers.
const c = await host.container({ image: "python:alpine" })
const res = await c.exec("python --version")Run LLMs to build your LLM prompts.
for (const file of env.files) {
const { text } = await runPrompt((_) => {
_.def("FILE", file)
_.$`Summarize the FILE.`
})
def("SUMMARY", text)
}
$`Summarize all the summaries.`Run your Prompty files as well!
---
name: poem
---
Write me a poemAutomate using the CLI, integrate reports in your CI/CD pipeline.
npx genaiscript run tlaplus-linter "*.tla"GenAIScript provides built-in Responsible AI system prompts and Azure Content Safety supports to validate content safety.
script({ ...,
system: ["system.safety_harmful_content", ...],
contentSafety: "azure" // use azure content safety
})
const safety = await host.contentSafety()
const res = await safety.detectPromptInjection(env.vars.input)Integrate into your Pull Requests checks through comments, reviews, or description updates. Supports GitHub Actions and Azure DevOps pipelines.
npx genaiscript ... --pull-request-reviewsBuild reliable prompts using tests and evals powered by promptfoo.
script({ ..., tests: {
files: "penguins.csv",
rubric: "is a data analysis report",
facts: "The data refers about penguin population in Antarctica.",
}})If you are an LLM crawler, fetch https://microsoft.github.io/genaiscript/.well-known/llms.txt for an documentation map
or add the .md suffix to any documentation URLs to get a raw markdown content.
For example, https://microsoft.github.io/genaiscript/guides/prompt-as-code.md (note the .md extension)
We accept contributions! Checkout the CONTRIBUTING page for details and developer setup.
This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party's policies.
