Backlog Tamer

Backlog Tamer is a Telegram bot that turns messy learning or project inputs into structured Notion backlog items.

It accepts links or notes in Telegram, uses an agent workflow to draft a project/task proposal, asks for approval or revision, and writes the approved result to Notion.

Capabilities

• Telegram bot input by polling locally or webhook in local/deployed mode.
• Agent-based intake triage for links, notes, and project ideas.
• Human approval flow before writing anything to Notion.
• Durable confirmation/update state using SQLite locally or Postgres/Supabase when deployed.
• AWS Lambda deployment with a lightweight webhook receiver and an SQS-backed worker.

Local Development

Install dependencies with uv, then create a .env file with the required Telegram, OpenAI, Notion, and database settings.

Run the polling bot:

make run

Run the local webhook server:

make webhook-dev PUBLIC_URL=https://your-ngrok-url

Useful local commands:

make test
make lint
make format-check
make webhook-info
make webhook-clear

Deployment

Deployment uses:

• AWS Lambda for the Telegram webhook and worker.
• SQS plus DLQ for queued Telegram updates.
• ECR for the Lambda container image.
• Secrets Manager for runtime secrets.
• Supabase/Postgres for durable state.

Build and push the image:

./scripts/build_and_push_image.sh

Apply infrastructure changes:

terraform -chdir=infra/terraform apply

After deployment, register the Telegram webhook to the Lambda Function URL with the configured webhook secret.

Automatic deployment runs from GitHub Actions after CI passes on main, and can also be triggered manually from the Deploy workflow. It expects a production GitHub environment with:

• AWS_REGION
• AWS_ROLE_ARN
• TF_STATE_BUCKET

For shared infrastructure state, copy infra/terraform/backend.hcl.example to infra/terraform/backend.hcl, set your S3 bucket name, then initialize once with:

terraform -chdir=infra/terraform init -backend-config=backend.hcl -migrate-state