Skip to main content
The Python SDK does the same as the API, with retries and streaming already handled. It works on Python 3.9 or later (AWS Lambda, Cloud Run and notebooks too) and its only dependency is httpx.

Your first document

create takes the same body as POST /documents as keyword arguments and waits until the document is finished. Responses are dictionaries with the same fields as the API.

Options

Review and edit

Read a file into Markdown

Free. Takes a path, bytes (with filename), a public URL, a file_… or a doc_…:

Follow the progress

Or without waiting: create_in_background(...) answers at once and wait(id) waits until it is finished.

Tools for your agent

doconda.tools() returns Doconda’s tools (create, upload, review, edit, list, get, report and an artifact’s data). Each has name, description, input_schema (JSON Schema) and run(args), so they work with any model that calls functions:
How to plug them in on each platform: Amazon Bedrock, Vertex AI, OpenAI, Claude and LangChain.

Webhooks

Errors

An error response from the API raises DocondaError, with status, code, errors, request_id and the message. 429s and 5xx are retried on their own (honouring Retry-After) before it is raised. See errors.

Reference