gogo-learn bot

overview

a Telegram bot that turns YouTube links into bite-sized learning snippets via Gemini, with SQLite persistence. Allow users to learn from videos anywhere, anytime, at their convinience.

Examples

• include cost report for each note *

setup

./setup.sh (or uv sync) → source .venv/bin/activate.

Before using this, you need to chat with BotFather on Telegram to create a new bot. Add the bot name and token in .env file.

You also need a Gemini dev API for the LLM part.

Finally, find your Telegram chat id by sending a message to the bot and observe the log and add that to allowed list.

.env file

TELEGRAM_BOT_USERNAME=
TELEGRAM_BOT_TOKEN=
GEMINI_API_KEY=
ALLOWED_CHAT_IDS=

Run bot locally

uv run python bot.py

Telegram usage

- /start → help text shown.
- Send a single short YouTube link (e.g. a 2-min video) → expect
"Watching" → "Note ready" → 5–15 snippet messages.
- Send a message with 3 mixed links (youtu.be + watch?v= + shorts) →
expect 3 sequential processing flows, each producing snippets.
- Write something along with the links to steer learning focus in the note generation process. Example '[url1] [url2] summarize key marketing tips in all these videos'.
- Send "hello" → expect "No YouTube links detected."
- /status → shows correct counts.

deploy to fly.io

one-time setup

fly auth login                          # if not already
fly launch --no-deploy --copy-config    # uses existing fly.toml; pick a unique app name if "gogolearn-bot" is taken
fly volumes create gogolearn_data --region sea --size 1   # SQLite lives at /data/gogolearn.db

Create a .env file in the project root with your secrets. Push secrets and deploy:

./deploy.sh --secrets   # loads .env → fly secrets set, then fly deploy

subsequent deploys

./deploy.sh             # fly deploy only (secrets already set)
./deploy.sh --secrets   # re-push secrets + deploy (e.g. after rotating keys)

pushing secrets without deploying

python set_fly_secrets.py   # reads .env and runs fly secrets set

operate

fly logs              # tail
fly status            # machine state
fly ssh console       # shell into the VM (DB at /data/gogolearn.db)
fly scale count 0     # stop/turn down the deployment
fly scale count 1     # this is a polling bot — keep exactly 1 machine

notes

• fly.toml mounts the gogolearn_data volume at /data and sets GOGOLEARN_DB=/data/gogolearn.db so the SQLite file survives deploys/restarts.
• No [http_service] block — the bot uses Telegram long-polling, so no public port is needed.
• Run a single machine. Multiple machines polling the same bot token will conflict.
• Never commit .env — it is listed in .gitignore.