ArgosBrain · For Shopify agent builders

Your Shopify agent
should remember
Dawn's section schema.

Persistent, deterministic memory for AI coding agents that build Shopify themes and apps. Liquid, JSON schema, section settings, and Dawn conventions — indexed once, recalled in sub-millisecond, at $0 per query.

01The problem

Shopify themes break coding agents.

You're building a Shopify agent — or using Claude Code / Cursor to write Liquid — and every session starts the same way: the agent re-reads sections/header.liquid, re-discovers the section's {% schema %} JSON, and re-guesses whether settings.color_scheme exists on the page object. Tokens burn; the agent still proposes snippets that violate Dawn conventions.

Liquid is not a well-served language in the tooling world. No Language Server Protocol. No canonical SCIP indexer. Tree-sitter support is patchy and version-pinned. So every agent that touches Shopify does regex + hope — and it shows.

If you're building Shopify-focused agents at scale, this is the silent tax on every output.

02What ArgosBrain does

Liquid as a first-class citizen.

ArgosBrain ships a vendored Liquid tree-sitter grammar with Dawn-aware indexing: sections, snippets, blocks, schema settings, and translation keys all land in the symbol graph as distinct node types. We wrote our own bindings so we aren't blocked by upstream grammar version pins.

  • Sections and snippets indexed as addressable symbols, with caller/callee edges across {% render %} / {% include %}.
  • Section {% schema %} JSON parsed and linked to the enclosing section — so resolve_member("header_section.settings.logo_width") returns a real answer in microseconds.
  • Dawn-style naming conventions learned automatically (naming_convention tool) so new sections the agent creates match the rest of the theme.
  • File-hash invalidation: when you edit sections/cart.liquid, only that section's chunks re-index. No full re-ingest.

All of this is local. No theme content leaves your machine. No LLM call on the read path.

03The proof

100% symbol recall on Dawn.

We ran the full LongMemCode methodology against Shopify's Dawn reference theme — 500 deterministic scenarios across completion, bug-fix, refactor, test-gen, feature-add, api-discovery, control-flow, and config tasks. Compared a regex / grep baseline against ArgosBrain's MCP adapter.

100%
ArgosBrain symbol recall
on Dawn corpus
24%
Regex / grep baseline
on same scenarios
500
Deterministic scenarios,
SCIP-gold + adversarial
$0
Per query on the
retrieval path

Methodology, scoring, and corpus generator are all open source. You can replicate the number on your own laptop in under an hour.

04Install

One line. Then your agent remembers.

curl -fsSL https://argosbrain.com/install.sh | sh
cd my-shopify-theme
argosbrain ingest .
# done — Claude Code, Cursor, Aider, any MCP agent now has memory

Ingestion walks sections/, snippets/, templates/, layout/, config/, and locales/. Parses Liquid + embedded JSON. Builds the symbol graph. Takes under a minute on a full Dawn fork.

After that, any MCP-compatible agent pointed at the argosbrain server gets seven new tools — recall, search, resolve_member, symbol_exists, list_symbols, naming_convention, introspect — and the memory persists across sessions, branches, and restarts.

05Who it's for

Agent builders shipping Shopify tooling.

If you're building a vertical agent for merchants — Shopify store generators, theme customizers, app scaffolders, onboarding bots that modify Liquid — ArgosBrain is the memory layer you'd otherwise have to build yourself. Badly. In six months. With a token bill to match.

If you're a solo Shopify developer using Claude Code or Cursor, it's the same engine, same install, same $0 per query. The solo tier is free forever.

Either way — tell us what you're building. We're actively prioritizing the Shopify surface area (Liquid translation-key resolution, theme-editor block graph, app extension manifests) and design-partner feedback moves it up the queue.

06Next