Project-specific guardrails
Findings are grounded in your real architecture model (layers, import graph, patterns, ADR decisions)—not generic lint rules.
Meridian
Architectural review while you write code.
Not a linter. A feedback loop for architecture—so boundaries stay intact as the codebase grows. Meridian catches dependency direction violations, boundary leaks, and pattern drift in real-time, grounded in your project’s architecture docs and ADRs.
Quiet by design: Low-confidence findings are dropped. Silence is a feature.
What it does
Defensible explanations
Each finding explains the rule, the evidence, the consequence, and a suggested fix—so teams can align and move fast.
Works where devs already are
Designed for Cursor / VS Code via MCP so it can sit directly in the developer workflow, not as a separate review ritual.
Your architecture stays yours
Architectural details and code structure shared during review are never persisted to remote servers, and no personally identifiable information is retained—so sensitive design decisions stay within your control.
Fully local AI inference
The LLM powering Meridian's analysis runs locally on the server and shares nothing with external entities, keeping your codebase and architectural intent air-gapped from third-party AI providers.
How it works
1. Scan
Builds an architecture model from documents and code-bases: layers, dependency direction, patterns, and ADR summaries.
2. Ground
Assembles a project-specific Resolving Architecture enhanced (from experience) system prompt and schema so findings are anchored in this codebase’s intent.
3. Review
Uses an embedded (in the server) AI provider and returns structured findings rendered inline in the IDE.
FAQ
Does it send my whole repo?
No. The local scanner builds the model. Reviews send only what’s needed for analysis.
Will it spam warnings?
No. Silence is a signal of health. Low-confidence findings are dropped.
How is billing handled?
Tokens are purchased with Stripe or Bitcoin. Pay-as-you-go with Lightning on the roadmap.
How do I get started?
Meridian is currently in Alpha being trained against models to ensure accuracy. Email us to get it once it's in Beta.
Keep architecture from becoming a post‑mortem topic.
Turn architecture intent into real-time guardrails.