Coordinate agents, run staged work with /dr-auto or tiny tasks with /dr-quick, and use Jev model advice. Get a clear account of the result and its evidence. 19 agents, 29 commands, 80 skills.
How to install → · Using an AI agent? Give it the repository URL and ask it to install Datarim
Datarim helps coordinate agents, continue work without constant prompting, and choose a suitable mode through Jev, and handle tiny tasks through /dr-quick. You set the goal; requirements, verification, and permission boundaries remain in force at every stage.
Planning, implementation, and review are assigned to the corresponding roles. The optional dr-orchestrate runner continues a terminal workflow from state snapshots and checks each action against the allowed policy.
/dr-plugin enable /path/to/plugins/dr-orchestratedr-orchestrate run --dry-run
The runner is installed separately and requires tmux. Unknown state or a disallowed action stops automatic continuation; new learned rules require confirmation.
Set up orchestration/dr-auto starts a new task or resumes an existing one, delegates its stages, and resolves small gaps from available evidence. Each stage reports its outcome and open limitations in plain language.
/dr-auto task description
On the full route, /dr-auto stops after successful compliance review (/dr-compliance) and its internal reflection. Archiving remains a separate action. Ambiguity and protected actions retain their required gates.
Start staged workJev classifies a prompt through the configured model API and recommends a working mode and model tier. In a Datarim project it also recommends catalogue artifacts. This helps match the task complexity to the executor. Use it with Datarim or independently.
jev doctor --agent=codex --api
Classification requires configured API access; advice does not automatically change the active session model. The command safety floor works without a key; when the API is unavailable, the session continues without advice.
Install and verify Jev/dr-quick handles a tiny self-contained edit or a read-only lookup. It skips the full requirements, planning, design, QA, and compliance stages, while retaining mandatory evidence for mutations.
/dr-quick "tiny task"
An edit needs a falsifiable success condition, live execution evidence, and delivery reachability from origin/main before a short archive. A lookup creates no task files and stays UNCERTIFIED: informational, without a certified completion claim. If scope grows, use the full cycle.
Choose the fast laneIn 4.2, reporting connects the outcome to the original request and evidence. If the explanation is unclear, /dr-explain restates it without executing work. Human Outcome Reporting.
Use your preferred reply language while creating documents in the project language. Independent language.replies and language.artifacts settings both default to en, including standalone reporting without Datarim. Configure languages.
Not just code — research, legal docs, content, project management. Datarim adapts its pipeline to task complexity: simple fix or enterprise feature.
Every task ends with a mandatory reflection — it runs at a passing /dr-compliance verdict (and /dr-archive regenerates it only if missing or stale) to analyze what worked and what didn't. The framework learns and improves itself through evolution proposals.
Create new skills, agents, and commands with /dr-addskill. Describe what you need in plain language — Datarim generates the artifact.
$ git clone https://github.com/Arcanada-one/datarim.git ~/src/datarim
$ cd ~/src/datarim && git checkout "$(git describe --tags --abbrev=0 --match 'v*')"
$ ./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 # see all commands
/dr-init my task # start workingSetup covers six choices: five settings and the source release. At a terminal the installer prompts for the settings (Enter takes the default) and reminds you to choose the release. Without one it installs nothing: it prints all six questions, the answer flags and a one-time --answers token; rerun with the user’s answers and that token. The questions, the flags and the report after the install are in the install guide (following INSTALL.md).
Datarim adapts to your role and workflow
From requirements to production with TDD, code review, and compliance.
Backlog, iterations, progress tracking, retrospectives.
Articles, social media, newsletters — with editorial review.
CI/CD, Docker, deployment automation, secret management.
Contracts, ToS, compliance docs with phased review.
Pre-launch checklist: DNS, SSL, analytics, social previews.
Datarim is developed as part of the Arcanada project — an ecosystem for building, orchestrating, and scaling AI agents.
Visit arcanada.ai →