Language
Pythonspeakerdeck.com
Main language for APIs, services and internal tools.
Hosting
Herokuspeakerdeck.com
Platform where the services ran, as of 2020.
Supplies API Gateway, SQS, SNS, S3, Lambda, RDS and Athena.
Database
PostgreSQLspeakerdeck.com
Main database, with full-text search, a read replica and partitioning.
Data fetching
Amazon SQSspeakerdeck.com
Queues that every service consumes to react to events.
Why: Packing every election and publishing rule into one synchronous API did not work, so they combined HTTP with pub/sub.
Redisspeakerdeck.com
In-memory data store listed in the 2020 stack.
Backend framework
Django REST Frameworkspeakerdeck.com
Builds the REST APIs of the microservices.
Second language, used for some of the APIs.
Loafergithub.com
Olist's open-source Python dispatcher that every service is built on to consume SQS.
Testing
Locustspeakerdeck.com
Load tests the APIs before peak dates, with Siege.
Code quality
pre-commitgithub.com
Runs Olist's custom lint hooks, published as the hulks project.
Search
Elasticsearchspeakerdeck.com
Search engine in the 2020 stack.
AI model
Claude Codegithub.com
Coding assistant developers use through Olist's AI gateway.
Error monitoring
Elastic APMspeakerdeck.com
Monitors API performance, alongside pgHero and django-silk.
OpenTelemetrygithub.com
Traces each background task run by the Django task worker.
WhatsApp and SMS
WhatsAppgithub.com
Receives incoming WhatsApp webhooks through a Django app.
In 2018 Olist described its architecture in four layers. APIs are RESTful entry points written in Django, Go or Apistar. Services hold the business rules, read and write the databases and consume SQS queues. Brokers group services around one goal, such as publishing to a given marketplace, and apps are the user-facing interfaces with no storage of their own.
Sources:speakerdeck.com
When a seller changes a price, the product has to be re-evaluated and republished to several channels. Cramming all those election and publishing rules into one synchronous API did not hold up, so Olist pairs HTTP with publish and subscribe.
Every service is built on Loafer, an open-source Python library from Olist that consumes Amazon SQS queues asynchronously and keeps message providers apart from consumers.
Sources:speakerdeck.com github.com
By 2020 Olist ran more than a hundred microservices on Python 3, Django REST Framework, Heroku, PostgreSQL and a long list of AWS services. To survive Black Friday, the team restructured database relations, turned on Postgres full-text search, added a read replica and partitioning, and brought in caching and async code.
They watched performance with Elastic APM, pgHero and django-silk, load tested with Siege and Locust, and released small libraries like a rapidjson renderer for Django REST Framework.
Sources:speakerdeck.com opensource.olist.com
Olist's public repositories show the same foundation today. Recent projects include an async worker for django-tasks-db with OpenTelemetry tracing and a Django app that receives WhatsApp webhooks.
The team also built olist-code, a local proxy that routes Claude Code and opencode through Olist's own AI gateway, with company login and per-model usage tracking.
Sources:github.com github.com github.com
Found on the homepage with the open Wappalyzer rules (HTML, scripts and headers). It shows what runs in the browser, not the whole backend.
Checked on 2026-10-29. Every tool links to the public source that shows it.
