stack.tools

Shopify

E-commerce platform powering millions of merchant storefronts and an AI merchant assistant (Sidekick).

Shopify runs one of the most public "AI-first" programs in tech: Tobi Lütke's April 2025 memo made reflexive AI usage a baseline expectation, and the company ships merchant-facing AI (Sidekick on Claude Sonnet via Vertex AI, the Shopify Magic suite, SimGym simulated shoppers) on top of a deliberately multi-provider model strategy. Their build-vs-buy philosophy is distinctive: buy frontier models and coding tools at scale (Cursor, Copilot, Claude Code), but build and open-source the orchestration and internal layers themselves (Roast, MCP-UI, a LibreChat fork for chat.shopify.io) and fine-tune open-source VLMs where API costs are prohibitive, with in-house LLM-judge evals and GRPO post-training for Sidekick.

shopify.com
Facts
19
Receipts
31
Layers
10

Sources last verified Jul 26, 2026

Models & Providers

4 facts

Claude Sonnet by Anthropic

Primary model behind Sidekick, the AI commerce assistant for merchants.

High confidence

Receipts · 2

Show quotes (2)
  • “The combination of Claude and Vertex AI helps us empower millions of merchants with our AI-enabled commerce assistant, Sidekick” — cloud.google.com, Feb 5, 2026
  • “Conversational artificial intelligence services” — help.shopify.com, Jul 2026

Source last verified Jul 26, 2026

OpenAI API by OpenAI

Additional model provider behind Sidekick's multi-provider routing.

High confidence

Receipts · 1

Show quotes (1)
  • “Personal data if input to or included in a response by Shopify Sidekick” — help.shopify.com, Jul 2026

Source last verified Jul 26, 2026

Qwen2VL 7B (fine-tuned) by Alibaba (open source)

Fine-tuned vision-language model powering the product-understanding pipeline.

High confidence

Receipts · 1

Show quotes (1)
  • “We have deployed three successive open-source models: LlaVA 1.5 7B, LLaMA 3.2 11B, and currently Qwen2VL 7B.” — shopify.engineering, Jul 16, 2025

Source last verified Jul 26, 2026

Shopify Magic by Shopify (in-house)

Free AI feature suite built on multiple leading model providers.

High confidence

Receipts · 1

Show quotes (1)
  • “built on a combination of Shopify's proprietary data and the world's leading large language models” — help.shopify.com, Jul 2026

Source last verified Jul 26, 2026

Inference & Serving

3 facts

Vertex AI by Google Cloud

Serves Claude models for Sidekick at merchant scale.

High confidence

Receipts · 1

Show quotes (1)
  • “The combination of Claude and Vertex AI helps us empower millions of merchants with our AI-enabled commerce assistant, Sidekick” — cloud.google.com, Feb 5, 2026

Source last verified Jul 26, 2026

Triton Inference Server by NVIDIA

Self-hosts fine-tuned vision models across the GPU fleet.

High confidence

Receipts · 2

Show quotes (2)
  • “Orchestrates model serving across our GPU fleet, handling request preprocessing, batching, and routing” — shopify.engineering, Jul 16, 2025
  • “40 million LLM calls daily, representing about 16 billion tokens inferred per day” — shopify.engineering, Jul 16, 2025

Source last verified Jul 26, 2026

GroqCloud by Groq

AI services provider whose exact workload is not publicly detailed.

Medium confidence

Receipts · 1

Show quotes (1)
  • “Artificial intelligence services” — help.shopify.com, Jul 2026

Source last verified Jul 26, 2026

Orchestration & Agents

2 facts

Roast by Shopify (open source)

Home-grown Ruby DSL for structured AI workflows.

High confidence

Receipts · 2

Show quotes (2)
  • “Non-determinism is the enemy of reliability.” — shopify.engineering, Jun 18, 2025
  • “Build complex AI-powered automation with simple, declarative Ruby syntax.” — github.com, Jul 2026

Source last verified Jul 26, 2026

Model Context Protocol (MCP) by Anthropic (open standard)

Exposes internal company data and storefronts to agents over MCP.

High confidence

Receipts · 2

Show quotes (2)
  • “Connect any AI assistant to real-time commerce data from Shopify stores” — shopify.dev, Jul 2026
  • “We make every single piece of data inside the company available” — firstround.com, Jul 15, 2025

Source last verified Jul 26, 2026

RAG & Vector Search

1 fact

Google Cloud Dataflow by Google Cloud

Streams real-time embedding pipelines feeding storefront semantic search.

High confidence

Receipts · 1

Show quotes (1)
  • “we use Google Cloud's streaming analytics service Dataflow to power these pipelines” — shopify.engineering, Oct 18, 2024

Source last verified Jul 26, 2026

Evals & Observability

1 fact

In-house LLM-judge eval platform (GTX + merchant simulator) by Shopify (in-house)

Evaluates Sidekick with calibrated LLM judges and a merchant simulator.

High confidence

Receipts · 2

Show quotes (2)
  • “we built an LLM-powered merchant simulator that captures the 'essence' or goals of real conversations” — shopify.engineering, Aug 26, 2025
  • “We moved away from carefully curated "golden" datasets toward Ground Truth Sets (GTX)” — shopify.engineering, Aug 26, 2025

Source last verified Jul 26, 2026

Fine-tuning & Training

1 fact

GRPO reinforcement-learning fine-tuning by Shopify (in-house)

Post-trains Sidekick's fine-tuned models with LLM judges as reward signals.

High confidence

Receipts · 2

Show quotes (2)
  • “we implemented Group Relative Policy Optimization (GRPO), a reinforcement learning approach that uses our LLM judges as reward signals” — shopify.engineering, Aug 26, 2025
  • “Fine-tuning an open-source model got us there for the common cases.” — shopify.engineering, Jun 15, 2026

Source last verified Jul 26, 2026

Data & Labeling

1 fact

Toloka

Supplies the human-annotated seed dataset calibrating automated training-data curation.

High confidence

Receipts · 2

Show quotes (2)
  • “We initially partnered with the Toloka team to produce a balanced dataset” — shopify.engineering, Jun 15, 2026
  • “If four independent models can't agree on a label, a human should make the call” — shopify.engineering, Jun 15, 2026

Source last verified Jul 26, 2026

Guardrails & AI Security

1 fact

Refusal-behavior training pipeline by Shopify (in-house)

Trains Sidekick to refuse impossible or out-of-scope requests.

High confidence

Receipts · 1

Show quotes (1)
  • “The model had never learned to say no.” — shopify.engineering, Jun 15, 2026

Source last verified Jul 26, 2026

AI Coding Tools

3 facts

Cursor by Anysphere

Standard AI code editor across engineering.

High confidence

Receipts · 2

Show quotes (2)
  • “I ordered 1,500 Cursor licenses last year and quickly had to procure another 1,500.” — firstround.com, Jul 15, 2025
  • “Developers have proxy, Copilot, Cursor, Claude code” — x.com, Apr 7, 2025

Source last verified Jul 26, 2026

GitHub Copilot by GitHub (Microsoft)

Longest-standing AI coding tool in the sanctioned developer toolkit.

High confidence

Receipts · 2

Show quotes (2)
  • “Pretty quickly, our adoption increased to 80% and GitHub started asking us how we did that” — firstround.com, Jul 15, 2025
  • “Developers have proxy, Copilot, Cursor, Claude code” — x.com, Apr 7, 2025

Source last verified Jul 26, 2026

Claude Code by Anthropic

Sanctioned agentic coding tool, also embedded as Roast's coding agent.

High confidence

Receipts · 2

Show quotes (2)
  • “There is chat.shopify.io, which we had for years now. Developers have proxy, Copilot, Cursor, Claude code” — techstartups.com, Apr 7, 2025
  • “Claude Code has transformed how we build internal tools at Shopify” — claude.com, Jul 2026

Source last verified Jul 26, 2026

Internal AI Productivity

2 facts

LibreChat (chat.shopify.io) by LibreChat (open source)

Runs the company-wide internal AI chat on an internal fork.

High confidence

Receipts · 2

Show quotes (2)
  • “Shopify runs an internal fork of librechat, and we merge most everything back.” — x.com, Jun 11, 2025
  • “Built using LibreChat (to which Shopify is a core contributor)” — firstround.com, Jul 15, 2025

Source last verified Jul 26, 2026

Reflexive AI usage mandate by Shopify (internal policy)

Company-wide policy making reflexive AI use mandatory in daily work.

High confidence

Receipts · 2

Show quotes (2)
  • “Reflexive AI usage is now a baseline expectation at Shopify” — x.com, Apr 7, 2025
  • “Before asking for more Headcount and resources, teams must demonstrate why they cannot get what they want done using AI” — techstartups.com, Apr 7, 2025

Source last verified Jul 26, 2026