Normik documentation
Normik is a quality gate for AI-generated code: every pull request is checked against your team's playbook, on your own API key. This documentation covers the whole product.
Core concepts
| Concept | What it is |
|---|---|
| Organization | Your tenant. Members, billing, the AI provider key and all content belong to it. Data is strictly isolated per organization. |
| Workspace | A folder inside the organization — typically one per client or product. Rules and skills placed in a workspace apply on top of the organization-wide ones. |
| Rule | One standard, written in plain language, with a severity (info / warning / error), glob scopes, tags and a version history. |
| Skill | A reusable how-to (markdown instructions) your coding agents can discover and load on demand via MCP. |
| Workspace token | A bearer token (nmk_…) that scopes CI jobs and MCP agents to a workspace — global rules are inherited automatically. |
| Analysis | One gate run: the diff, the rules applied, the violations found. Feeds the compliance dashboard. |
The two enforcement surfaces
The same playbook is enforced in two places, and they reinforce each other:
- At write time — coding agents (Claude Code or any MCP client) load your rules while they work, self-check their diff and fix violations before the commit.See MCP & agents.
- At merge time — a CI job checks the pull request diff and blocks the merge on error-severity violations.See CI gate.
Where to start
- Quickstart — from signup to a gated pull request in five steps.
- Playbook & rules — writing rules that a model can actually enforce.
- Security & data — what happens to your code and keys (short answer: the code isn't stored, the keys are encrypted).