Platform · Component ② · Parts (Utilities)

Proven shared components,
plugged in via pip

standarda-core is a package that bundles the functionality every AX project needs — LLM integration, Google APIs, document parsing — into proven components. Plug it in with a single pip, and when you improve a part in one place, every project inherits that improvement instantly.

🧩 Proven shared components 📌 Pinned version-tag installs ♻️ Improve once → applied everywhere
v0.18Current release (continuously expanding)
5 LLMsClaude · OpenAI · Gemini · OpenRouter · embeddings
Google APIsGmail · Sheets · Drive · Docs clients
Document parsingPDF · Excel Vision + text extraction
Analogy · Car parts

You don't re-machine the engine and wheels every time

Just as a carmaker shares proven engines, transmissions, and electronics across many models, Dawell Service builds core parts like LLM calls, Google APIs, and document parsing well once and reuses them across every project. Improve a part in one place, and every car (project) that uses it gets better together.

  • Reuse proven parts → never build the same bug twice
  • Pin versions with tags → managed safely per project
  • When a new LLM or feature arrives, swap just the part and it applies everywhere
Car parts illustration — standarda-core
standarda-core = the proven shared components you plug into a project
The Parts

The parts inside

The functionality AX agents need over and over, collected in a proven form.

🤖

LLM integration

Switch between Claude, OpenAI, Gemini, and OpenRouter with an environment-variable toggle, plus OpenAI embeddings. Change models with no code changes.

💸

Caching · cost tracking

Prompt-cache control and cumulative token usage with cost estimation cut LLM call costs and keep them transparent.

🔁

Agent tool loop

A LangGraph-based tool execution loop — the standard flow for an agent calling tools in sequence, provided as a part.

📮

Google API clients

Gmail, Sheets, Drive, and Docs clients with standard authentication. Connect safely to the customer's Google Workspace.

📄

Document parsing

Analyze PDFs and Excel with Vision or extract them as text. Pull the values you need out of documents, structured.

🧾

Response parsing

Utilities that parse LLM responses into reliable JSON — keeping automation pipelines from breaking.

Principle ①

Single source of truth

The same functionality lives in exactly one place. Fix it here and it applies to every project.

Principle ②

Standard authentication

Google and LLM auth in a standardized way — no reinventing authentication per project.

Principle ③

Vision-First

For documents with complex layouts, we understand them with Vision instead of OCR to raise accuracy.

How It Works

Plug in a part, and improve it together

Install pinned to a version tag, and when a part gets better, just bump the tag to roll it out everywhere.

STEP 1

Pinned install

Each project installs the part at a specific version tag (e.g. @v0.18.0) — preventing unintended changes from propagating.

STEP 2

Improve the part

New features and bug fixes go into core just once. After review (PR), they ship as a new version tag.

STEP 3

Roll out everywhere

When each project bumps its tag, the improved part spreads safely to every case.

Python package (pip) git tag-pinned install Claude (Anthropic) OpenAI · Gemini · OpenRouter LangChain · LangGraph Gmail · Sheets · Drive · Docs API Vision document parsing Token cost tracking
Explore More

Explore the platform