Product
- What exactly does Normik check?
- Your playbook — rules your team wrote in plain language, each with a severity (info, warning, error) and optional glob scopes. Normik checks the pull-request diff against the rules whose scopes match the changed files. It does not do generic “best practices” review: if you didn't write the rule, it isn't checked.
- How is this different from a generic AI reviewer like CodeRabbit or Copilot?
- Generic reviewers tell you what the model thinks about your code. Normik enforces what your team decided: explicit, versioned rules with severities you chose, where error-level violations actually block the merge. See our detailed comparisons on the /vs page.
- Which languages and frameworks are supported?
- Any language a modern LLM can read — rules are written in plain language and judged on the diff, not parsed by language-specific analyzers. Use glob scopes (src/**, terraform/**) to target stacks precisely.
- Can I start with an empty playbook?
- You don't have to. Click Bootstrap from repo on the Playbook page (or point a connected agent at the bootstrap_playbook capability) and pick a handful of representative files — Normik derives a first set of specific, verifiable rules from your existing conventions. They land in a review queue — you approve what actually reflects your standards.
- What are skills?
- Reusable how-tos (markdown instructions) your coding agents discover and load on demand via MCP — release checklists, migration procedures, review guides. Same hierarchy, versioning and governance as rules, with usage analytics.
- Does Normik post comments on my pull requests?
- Yes, on both — and always from your own pipeline, never from a bot we operate. On GitHub Actions it works out of the box: the CI job posts a violation summary with its own GITHUB_TOKEN. On GitLab it takes one variable: the built-in CI_JOB_TOKEN is not allowed to write MR notes, so the job posts the note with a project (or personal) access token you provide as NORMIK_GITLAB_TOKEN. Skip it and nothing breaks — the verdict stays in the job log and the dashboard.
CI gate
- What happens if Normik or my AI provider goes down?
- The gate fails open: the check passes and explains why it couldn't run. A quality gate should catch bad code — never hold your deliveries hostage to a third-party incident. Blocking is strictly reserved for error-severity violations.
- What blocks a merge, exactly?
- Only violations of rules with severity “error” — the gate returns HTTP 422 and the CI job fails. Warning- and info-level violations are reported without blocking. You choose each rule's severity.
- Do I need to install a GitHub App or configure webhooks?
- No. The integration is one CI job that pipes the git diff to Normik over HTTPS with a workspace token. Nothing else touches your repository, and there's no third-party app authorization to grant.
- Can the AI hallucinate violations?
- Guardrails bound it: Normik only accepts violations that cite an existing rule and a line actually added in the diff — anything else is dropped. Every violation can also be marked as a false positive with one click; confirmed ones are fed back into the next analysis on that rule and file so the same call doesn't keep repeating. The dashboard's Rule reliability card tracks the false-positive rate per rule and flags the consistently noisy ones so you know which to reword or scope tighter.
Agents & MCP
- How do coding agents like Claude Code use my rules?
- Normik exposes a remote MCP server. Agents connect with a URL and a workspace token, load the playbook while they write, self-check their diff before committing, and report results — the dashboard shows what they fixed pre-commit.
- Do I need to install an MCP package on every laptop?
- No — the server is remote (Streamable HTTP). One `claude mcp add` command or a few lines in .mcp.json, with the token as the only credential. Nothing to version or update locally.
- Can agents change my playbook?
- They can propose — rules and skills suggested by agents are flagged AI-generated and wait in a review queue until a human approves them (validation is on by default). Nothing enters the enforced playbook without your sign-off.
- My tooling doesn't support MCP. Am I stuck?
- No — export your playbook as an AGENTS.md or CLAUDE.md file (one click, deterministic) and commit it to the repo. Any tool that reads context files ships with your standards.
Security & privacy
- Does Normik store my source code?
- No. The diff is analyzed in memory on your own API key and discarded — never written to disk or database. What's stored: your playbook, analysis outcomes (rules checked, violations with path/line), usage counts and the audit log.
- What is BYOK and why does it matter?
- Bring Your Own Key: analyses run on your own API key from Anthropic, OpenAI, Mistral or Google. You keep control of the model, the data path and the token bill — and because Normik doesn't resell inference, analyses are unlimited on every plan.
- How are my keys and tokens protected?
- AI provider keys are encrypted at rest with AES-256-GCM and never displayed after saving. Workspace tokens are stored as SHA-256 hashes, shown once at creation, revocable instantly. Every query is scoped to your organization from the session — never from request parameters.
- Is Normik SOC 2 certified or self-hostable?
- Not today — we'd rather say it plainly. If SOC 2 or self-hosting are hard requirements now, an enterprise platform like Packmind may fit better; our comparison page says exactly that.
Plans & billing
- How is Normik priced?
- Free (1 workspace), Team at €39/month (3 workspaces, 10 members), Agency at €149/month (unlimited client workspaces, up to 50 members, governance, audit log), Enterprise at €349/month (120 members included, extra members at €2.50/month each, priority support). Paid plans start with a 14-day trial, no card needed for Free.
- Why are analyses unlimited?
- Because they run on your key, your provider bills the tokens — we have no marginal cost to meter. You pay Normik for organization: workspaces, members, governance. The dashboard shows your exact token spend so the BYOK bill is never a surprise.
- What happens if I downgrade?
- Limits apply at creation time only — you never lose access to existing workspaces or data. You just can't create beyond the new plan's limit.