OpenTremor

For the enterprise

The Engineering Standards Platform for the Age of AI Agents.

Turn the decisions in your ADRs and agent instructions into rules, and apply them to every product: its repositories, pull and merge requests, and code on its default branch.

  • GitHub, GitLab or any CI
  • Self-hosted or Cloud
  • Open source

100 free analyses on Cloud. Or self-host it: unlimited, forever. No sales call either way.

Agents now write pull requests faster than your team can read them. Reviewer time didn't grow with them.

Define what matters once and apply it to every PR. Your engineers spend their time on the changes that need them.

How it works

One set of standards for every product. Security, policy and style are just rules inside it.

You are not buying a code reviewer, a security scanner, and a governance platform. One review runs on each product's code, on the default branch and on every pull request, and your rules decide what it looks for.

  1. 1

    Define your standards

    Approve the rules drafted from your ADRs and agent instructions, or write your own. A rule is a Markdown paragraph written the way you'd explain it to a new engineer. Apply it everywhere or to one team, product or repository, alongside the built-in security rules.

    Each rule is a Markdown paragraph, drafted from your ADRs or written by hand.

    ## Backend standards

    - **API-004 (CRITICAL/HIGH):** a new endpoint reads the database without filtering on the caller's tenant, so one customer can see another's data.

  2. 2

    Review existing code and every change

    Register your organization's GitHub App in one click, or connect a self-managed GitLab group, and group repositories into products. OpenTremor scans each default branch, then reviews every pull or merge request. Any other CI can call the merge gate.

    Scan each product's existing code, then review every pull or merge request. Low-confidence findings wait for a reviewer.

    • Each file goes to the pack for its language or tool.
    • A first scan covers the code already on the default branch.
    • Findings land in one PR comment. A commit status holds the merge.
    • The owning team hears about it in Slack, Teams, email, or a webhook.
  3. 3

    Decide once

    A reviewer acknowledges, suppresses or marks a finding as a false positive in the app or on the PR. The decision carries over to the next scan and counts against the rule that raised the finding.

    Dismiss a finding once, and the next scan remembers.

    AcknowledgeSuppressFalse positive
See it work: nine user stories, recorded in the productA pull request review, a scan of existing code, rules from ADRs, goals, triage, a coding agent, prompt injection and the support assistant.

Covered today: Python, TypeScript and npm dependencies; GitHub Actions and GitLab CI; Dockerfiles, Docker Compose, Kubernetes, Helm, Ansible and Terraform, each handled by a pack that understands the format. Any other file goes to the generic pack. It does not parse the format: it looks for secrets, unsafe calls and similar problems, with medium confidence at most. Each pack is a plain Python package, and you can register your own. Code, CI, containers and infrastructure each have their own pack. Other files get broader checks. See packs

Rules from your own documents

Your engineering decisions already exist. They just aren't checked.

OpenTremor reads the ADRs and agent instruction files in your repositories and drafts the rules they imply. Nothing runs until a person approves the rules. A file that tries to steer the review bot is held back and flagged.

Read the docs

adr/0007-tenant-scoping.md

## Decision

Every query on a tenant collection filters on the caller's organization. A request never reads another customer's data.

Proposed rule

**API-004 (CRITICAL/HIGH):** a new endpoint reads the database without filtering on the caller's tenant, so one customer can see another's data.

ApproveEditReject

Feedback loop

Reviewer decisions improve the rules, not just the findings.

Every decision counts against the rule that raised the finding. When reviewers keep rejecting a rule, OpenTremor proposes an edit. A person applies it. Findings are never suppressed automatically.

Read the docs

Rule health

API-004

  • Findings this month8
  • Marked false positive6
  • Fixed2

Proposed edit

**API-004 (CRITICAL/HIGH):** a new endpoint reads the database without filtering on the caller's tenant. Platform-admin endpoints that already check the admin role are out of scope.

ApplyReject

Already running Semgrep, Checkov, SonarQube or GitHub Advanced Security?

Keep them. OpenTremor reviews what traditional scanners cannot express.

Scanners are good at known-bad syntax. Most of your standards are not about syntax. They live in a wiki, and today a person checks them by hand on every PR.

QuestionPattern-based scannersOpenTremor
Where rules come fromWritten by hand in the tool's languageDrafted from your ADRs and agent instruction files, then approved by a personDrafted from your ADRs and agent instructions
A rule isA pattern or policy written in the tool's query language or codeA Markdown paragraph that anyone on the team can write and review in a PR
It catchesKnown-bad constructs: an injectable query, a wildcard permission, a hardcoded keyIntent against your conventions: a query missing tenant scoping, a blocking call in an async path, a library your team has bannedYour engineering conventions, not just known-bad syntax
When unsureAlways gives a pass or failReports its confidence. Low-confidence findings keep the PR check pending and notify a person.Waits for a person instead of guessing
Your decisionsDismissed one finding at a timeRecorded against the rule. When reviewers keep rejecting a rule, OpenTremor proposes an edit, and a person applies it.
Agent-written codeMostly out of scopeChecks every added line for text aimed at your review bots, not at you

They work well side by side. OpenTremor isn't a linter or a SAST replacement. The Python pack says so itself: ruff and mypy already cover style, so it doesn't.

AI agents

Agents write your code. They can run OpenTremor too.

Agents as code authors

Their changes meet the same standards.

An agent's pull request is reviewed against your rules like anyone else's. OpenTremor also checks every added line for text aimed at AI reviewers and agents rather than at people. This check uses no LLM and is free on every plan.

Invisible UnicodeAgent files from a forkPrompts in comments

Agents as operators

They work in OpenTremor as you.

Over MCP, an agent runs analyses, triages findings and writes rules with the permissions of the person who connected it, and nothing more. The audit log names the token behind each action.

Try OpenTremor

Start today, without a procurement process.

Cloud

100 free analyses

  • Create an organization, then connect GitHub or your CI
  • Every pack and every feature. Nothing is gated by plan.
  • After that, a dedicated instance with usage-based pricing

Self-hosted

Unlimited, forever

  • Open source under the AGPL v3
  • Docker Compose on one VM, or Helm on Kubernetes
  • Your LLM key, your network, your data retention

Need a support contract, a dedicated instance, or a pack built for your internal formats? See Enterprise & Services

Stop reviewing everything. Review what matters.