@doconda/sdk weekly downloads statistics

@doconda/sdk

Generate, review and edit Word (DOCX), PowerPoint (PPTX), Excel (XLSX) and PDF documents from your AI agent or chatbot, with one API call. Editable files with real Word styles, reviewed before they are delivered.

Use it when your agent has to hand the user a document and you don't want to maintain python-docx, docx.js, pptxgenjs or templates. Works with the Vercel AI SDK, OpenAI, Azure OpenAI, Claude, Amazon Bedrock, Vertex AI, Gemini, Mistral, LangChain, Mastra and any model with tool calling.

npm install @doconda/sdk

import { Doconda } from "@doconda/sdk"

const doconda = new Doconda() // reads DOCONDA_API_KEY
const doc = await doconda.documents.create({ format: "docx", prompt: "Report on the housing law" })
console.log(doc.outputs) // links to the Word file and the PDF

Error messages and report texts come in English. For Spanish, pass new Doconda({ language: "es" }) (sent as Accept-Language). Codes (e.code) never change with the language.

Any file (PDF, Word, PowerPoint, Excel, images) to Markdown for your model, free up to 1,000 pages a month, then €0.01 every 5 pages. It takes a local path, a URL, bytes or a file_… id:

const { markdown } = await doconda.extract("proposal.pdf") // a file on this machine

As tools for your agent

doconda.tools() returns ready-made tools (name, description, JSON Schema, run) to create, review and edit documents. If the API answers with an error, run doesn't throw: it returns { error: "<code>", message, fields? } so the model can act on it.

import { anthropic } from "@ai-sdk/anthropic"
import { generateText, isStepCount, jsonSchema, tool } from "ai"
import { Doconda } from "@doconda/sdk"

const tools = Object.fromEntries(
  new Doconda().tools().map((t) => [
    t.name,
    tool({ description: t.description, inputSchema: jsonSchema(t.inputSchema), execute: (input) => t.run(input as Record<string, unknown>) }),
  ]),
)

await generateText({
  model: anthropic("claude-sonnet-5-5"),
  tools,
  stopWhen: isStepCount(5),
  prompt: "Create a 5-slide deck on our quarterly results",
})

Why Doconda

• Editable files: real Word styles, slide layouts and Excel formulas, not images or HTML exports.
• Reviewed before delivery: every file is checked and repaired; problems are reported, never hidden.
• Always a PDF and a preview of each page, ready to show in your chat UI.
• Works with any model that calls tools, and runs in the EU.

Examples

• Excel from an Amazon Bedrock agent
• A Vercel AI SDK chatbot that returns a Word file
• PowerPoint from Gemini on Vertex AI
• Review and fix a Word file before sending it
• More, one per stack (Bedrock, Vertex AI, Azure OpenAI, OpenAI, Claude, LangGraph, Strands, Mastra): B-C-Software/doconda-examples

Also in Python: doconda · Guides for every stack: https://docs.doconda.com/en/agents/introduction · MCP server: @doconda/mcp · Get an API key: https://doconda.com