document-analyzer

Document Analyzer

Hand it a PDF or DOCX as a base64-encoded file part in the A2A message, plus a prompt, and the agent extracts text and answers questions about the document. Agno + OpenRouter (openai/gpt-oss-120b). Uses pypdf for PDFs and python-docx for Word.

Setup

export OPENROUTER_API_KEY=<get one at https://openrouter.ai/keys>
uv sync --extra agents

Run

uv run examples/document-analyzer/document_analyzer.py
# http://localhost:3773

Talk to it

The agent reads parts[].kind == "file" for the document and parts[].kind == "text" for the prompt. A2A bodies carry files as base64.

With AUTH__ENABLED=false, no document attached:

curl -sS http://localhost:3773/ \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"message/send","id":"00000000-0000-0000-0000-000000000004","params":{"message":{"role":"user","parts":[{"kind":"text","text":"What is this about?"}],"kind":"message","messageId":"00000000-0000-0000-0000-000000000001","contextId":"00000000-0000-0000-0000-000000000002","taskId":"00000000-0000-0000-0000-000000000003"},"configuration":{"acceptedOutputModes":["application/json"]}}}'
# → "No valid document found in the messages."

To actually analyse a document, append a file part with base64 bytes alongside the text part. The full file-handling shape lives in docs/FILE_HANDLING_&_UPLOADS.md. With auth on, sign each body with the agent's DID key — see docs/AUTH.md.