langchain-doccheap

LangChain tools for doc.cheap: an agent gives them a photo or scan of a passport, national ID card or driving licence and gets back the printed fields – name, date of birth, document number, expiry, the machine-readable zone – as JSON.

One recognised document costs one credit, $0.01. Registering gives 100 free documents every month. An unreadable image, an empty frame or an unsupported document type costs nothing, and every answer says whether it was billed.

The package needs Python 3.10 or later. It is built on the doc-cheap client and langchain-core 1.x.

Installation

pip install langchain-doccheap

The tools

Every tool returns the API's answer as a JSON string. An API error – an unknown scan, a used-up rate limit, a wrong key – comes back to the model as a short text such as doc.cheap error: 404 scan_not_found: … instead of stopping the agent; pass handle_tool_error=False to have it raised as a ToolException.

Quick start

from langchain_doccheap import DocCheapScanDocument

scan = DocCheapScanDocument(api_key="sk_sandbox_public")
print(scan.invoke({"image_path": "passport.jpg"}))

With an agent, take all five tools from the toolkit. They share one client and its connections:

from langchain.agents import create_agent
from langchain_doccheap import DocCheapToolkit

tools = DocCheapToolkit().get_tools()  # reads DOC_CHEAP_API_KEY
agent = create_agent("openai:gpt-4.1-mini", tools)
agent.invoke({"messages": [{"role": "user", "content": "Read the passport in ./passport.jpg"}]})

DocCheapToolkit(include_delete=False) leaves out the tool that deletes stored results.

API keys

• Live key – sk_live_…, created in the doc.cheap cabinet.
• Sandbox key – sk_sandbox_public works without an account. It gives 10 free recognised documents per address in all, and at most 10 requests per address an hour, whatever their answer.

Without api_key the tools read the key from the DOC_CHEAP_API_KEY environment variable. base_url (or DOC_CHEAP_API_BASE) points them at another address. To control timeouts, proxies or connection pooling, build a doc_cheap.Client yourself and pass it as client=; the tools leave it open.

Stored results. A result is stored only when the scan was made with a live key under a non-zero retention window. Under the sandbox key nothing is stored, so getting or deleting a scan answers scan_not_found.

Resources

• doc.cheap documentation
• Your first recognition
• API reference
• The doc-cheap Python client

License

MIT