Integrating with an AI agent

Write an integration with the AiGenda API: upload a meeting recording, process it, get the protocol.

Documentation: https://docs.aigenda.tech/llms-full.txt. Read it in full before you write any code.

Where the recording is: <file on disk / URL / YouTube / meeting bot>
Language and stack: <Python 3.12 + requests / Node.js 20 / ...>
Where to put the result: <protocol.md file / our database / ...>

Required rules from the documentation:
1. Token: POST /api/token/ with the fields username (email) and password. Header Authorization: Bearer <access>.
2. The recording title ends with the file extension (.mp3, .m4a, .wav, .ogg, .webm, .mp4).
3. The file is uploaded in parts with PUT to the links from the url array; the ETag of each part is passed to complete_upload with its quotes.
4. Before classify, wait for is_uploaded: true in GET .../info/.
5. Poll the result every 15 seconds: 200 + classified: true + success: true = done; 200 + success: false, or 500 for more than 3 minutes = failed; 404 = processing was not started.
6. Do not blindly retry the POST that creates a recording: every call creates a new recording.
7. Take the email and password from the AIGENDA_EMAIL and AIGENDA_PASSWORD environment variables.
8. Meeting bot: always send meeting_service (telemost, zoom, or google_meet) to POST /api/space/{space_id}/meeting/, otherwise the bot is sent as if to Google Meet. Processing of a bot recording starts on its own.

Start with a plan, then the code, then instructions for running it.