✉️ Ara

Ara is an AI-powered email communication assistant. This repo contains the backend APIs, based on staart/api.

Production API base URL: https://eiva-api.o15y.com

APIs

Apart from Staart API endpoints that let you create and manage accounts and billing, Ara also has some email communication-specific APIs that developers can use:

• POST /v1/api/classify classifies text into actions
• POST /v1/api/parse-email parses a raw email into structured data
• POST /v1/api/smart-tokenize tokenizes text to actionable sentences
• POST /v1/api/perform-action takes an email and processes it
• POST /v1/api/read-receipt saves a read receipt for an email
• POST /v1/api/meeting-page/:username/:id returns public meeting details
• POST /v1/api/confirm-meeting/:organizaionId/:meetingId confirms a meeting
• POST /v1/api/track/:index tracks a usage event

CRUD endpoints:

• /v1/organizations/:id/location
• /v1/organizations/:id/meetings

All API endpoints require an API key or access token; both can be generated using APIs or the webapp. Parameters are available in api/index.ts.

👩‍💻 Development

Known issues

JSON

⚠️ FIX AVAILABLE: The fix for this is available in Prisma beta 5

String in Prisma schema doesn't allow larger fields, so you have to manually convert the following fields to LONGTEXT from VARCHAR after generating tables. For details, see prisma/migrate#116

• "incoming-emails".from
• "incoming-emails".to
• "incoming-emails".cc
• "incoming-emails".logs
• locations.data
• meetings.proposedTimes
• meetings.guests

You might also have to make some of these changes after running npx prisma migrate up --experimental because it would overwrite the database schema.

Database hanging

Sometimes, the Prisma database hangs (blocks the main thread) during the initial tests. The trick is to make sure NODE_ENV = "development" is in env. I think this has something to do with line 7 of src/_staart/helpers/prisma.ts.

Differences in _staart

• helpers/mail.ts supports more parameters
• init-tests.ts has ===== line

Resources

AWS S3 bucket

The ara-assistant-incoming-emails bucket is used to store incoming emails to Ara as plain text files that can be fetched and parsed by this API. It's hosted in the eu-central-1 region and has limited, non-public access. An example of such an email is available in content/2mgh53qnuk650do2k3qlb5pv27obrl22uga8de01 and is called using /v1/api/webhooks/inbound/email/OBJECT_ID?secret=SECRET as explained below.

AWS Lambda function

This serverless function EmailForwarder in invoked from AWS S3, when a new object is added to the emails bucket. The source code is available in content/mail-forwarder.js and is hosted in the eu-central-1 region with the Node.js 12.x runtime, 128 MB memory limit, and 1 minute execution timeout.

Redis

Key-value storage Redis is used for JWT cache invalidation and MySQL query caching. For production, a self-hosted Redis instance is used using Caprover, available only to other Caprover applications. For development, a local redis-server with default configuration will do.

Server

A dedicated server is used for deploying the EIVA service. This uses AWS Lightsail which provides an easy-to-use interface for AWS EC2.

Instance details:

• Region: Frankfurt (eu-central-1)
• RAM: 4 GB
• Processor: 2 vCPUs
• Storage: 80 GB SSD
• Public IP address: 18.195.203.61
• Billing period: May 21, 2020–present

ElasticSearch

A dedicated ElasticSearch instance is used to store server logs, and more importantly track usage events using the /v1/api/track/:index API endpoint. This data will be used for analytics about pages, time on site, etc.

Instance details:

• Elasticsearch version: 7.4
• Instance type: r5.large.elasticsearch
• Number of nodes: 1
• Data nodes storage type: EBS
• EBS volume type: General Purpose (SSD)
• EBS volume size: 10 GiB
• Billing period: May 21, 2020–present

Environment variables

These environment variables (with the exception of DATABASE_URL) can be set in the .env file, or available as environment variables in the process using the Caprover UI or Dockerfile.

General @staart/api

• PORT is 7007 in development and 80 on production
• BASE_URL = https://api.myeiva.com
• FRONTEND_URL = https://myeiva.com
• REDIS_URL is redis://:KvEqnrLZJhGGEuNHNMcgG3sH@srv-captain--redis:6379 in production and not required for development if you have a local Redis instance running using redis-server
• DATABASE_URL = mysql://USER:PASSWORD@HOST:PORT/DATABASE is set in .env in the ./prisma dir

AWS SES emails

This is used for both transactional emails (like password resets) using SES_EMAIL and for sending emails from Ara (like meet-anand@mail.assistant.com)

• SES_EMAIL is the email to send emails from (noreply@mail.myeiva.com)
• SES_REGION is the AWS region (us-east-1)
• SES_ACCESS is the AWS access key for SES
• SES_SECRET is the AWS secret key for SES

Stripe

• STRIPE_SECRET_KEY is the test key for development and live key for production
• STRIPE_PRODUCT_ID is the product ID for Ara (prod_HAdEYq2FQrjVSd for testing)

Clearbit

Clearbit is used to find information about guests, emails, and more. There are multiple API keys that are rotated. The secret API keys are stored in variables like:

• CLEARBIT_SECRET_KEY_1
• CLEARBIT_SECRET_KEY_2
• CLEARBIT_SECRET_KEY_3...
• CLEARBIT_SECRET_KEY_10 (up to 10 API keys are supported)

ElasticSearch

• AWS_ELASTIC_ACCESS_KEY is the AWS access key for ElasticSearch
• AWS_ELASTIC_SECRET_KEY is the AWS secret key for ElasticSearch
• AWS_ELASTIC_HOST is the AWS ElasticSearch endpoint

Webhook

For incoming emails, the /v1/webhooks/inbound/email/OBJECT_ID?secret=SECRET URL is used. The following is a real example for the URL that will work locally:

http://localhost:7001/v1/webhooks/inbound/email/2mgh53qnuk650do2k3qlb5pv27obrl22uga8de01?secret=ebbb11de0c0400bf869bd48537ab56676715ec78173191ed377b0cae9a3eb6d0

To generate the secret, sign the object ID using HMAC with SHA-256 and the secret. The secret is stored as the environment variable INCOMING_EMAIL_WEBHOOK_SECRET as defined above. In Node.js, you can generate it using:

const { createHmac } = require("crypto");
createHmac("sha256", INCOMING_EMAIL_WEBHOOK_SECRET)
  .update(objectId)
  .digest("hex");

Deployment

The environment variable PORT is used to run the app on a specific port. In production, this is 3000.

The following command is used to redirect port 80 to 3000 (source):

sudo iptables -t nat -A PREROUTING -i eth0 -p tcp --dport 80 -j REDIRECT --to-port 3000

PM2 is used to run the app, and updating it is done using ./update.sh.

📄 License

• Code: Server Side Public License
• Logo and assets: CC BY 4.0
• GeoLite2: CC BY 4.0
• GeoNames: CC BY 3.0