What is Datarim? An AI Workflow Framework for Claude Code

When you start using AI assistants for real work — not just one-off questions, but sustained project execution — you quickly hit a wall. The AI forgets context between sessions. There's no structure to guide multi-step work. Quality varies wildly. And there's no learning from one task to the next.

Datarim solves this. It's a framework that provides structure, memory, and quality control for AI-assisted project work — built for Claude Code, and one project install now also serves Codex CLI and Cursor.

The Pipeline

Every task in Datarim follows a pipeline:

init → prd → plan → design → do → qa → compliance → archive

But not every task needs all 8 stages. A simple bug fix goes: init → do → archive (reflection runs at /dr-compliance, regenerated by archive only if missing). A complex feature gets the full treatment with product requirements, architectural design, multi-layer QA, and compliance hardening.

Datarim routes automatically based on complexity levels (L1-L4). You focus on the work; the framework manages the process.

19 Specialized Agents

Each stage is powered by specialized agents — planner, architect, developer, reviewer, compliance officer, strategist, writer, editor, DevOps, SRE, security analyst, and more. They each have a focused role and collaborate through the pipeline.

For critical decisions, Consilium assembles a custom panel of agents to debate and synthesize a recommendation — like an architectural review board, but AI-powered.

Not Just for Code

Datarim works for any iterative work:

• Software development — from requirements to production with TDD
• Research — literature review, methodology, writing, fact-checking
• Technical documentation — API docs, architecture decisions, guides
• Content creation — articles, social media posts, with editorial review
• Project management — backlog, iterations, retrospectives

Self-Evolving

After every task, a mandatory reflection runs at a passing /dr-compliance verdict (and /dr-archive regenerates it if it is missing) — what worked, what didn't, and concrete proposals for improving the framework itself. Datarim literally learns from its own experience, and reflection cannot be skipped.

Get Started

git clone https://github.com/Arcanada-one/datarim.git ~/src/datarim
cd ~/src/datarim && ./install.sh --project /absolute/path/to/project   # no answer flags: asks the questions, or prints them for the user
cd /absolute/path/to/project && claude
/dr-help

At a terminal the installer asks its questions itself; without one it installs nothing and prints the questions, the flag for each answer and a one-time answers token for the rerun. The questions and flags are in the install guide.

Datarim is open source (MIT License) and part of the Arcanada Ecosystem.