Coval. The reliability layer for AI coding.
Coval makes AI coding agents trustworthy in your repo. It catches the APIs they invent, keeps their context grounded in what your code actually is, and keeps your codebase clean enough for them to keep working in — so the code your agent ships is correct, not just plausible. It's free to use and plugs into the tools you already use.
- HIGHauth/session.tsDuplicate of token-refresh in api/client.ts (0.94)
- HIGHlib/payments.tsHallucinated import: `stripe/v3-helpers` does not exist
- MEDutils/date.tsDead code — `formatLegacy()` has 0 callers
- MEDcore/graph.tsHub function: 11 callers, complexity 34 — split candidate
- LOWui/Modal.tsxPossibly misplaced — belongs in components/overlay
A feedback loop that keeps agents honest
AI agents drift and hallucinate because they don't have accurate, lasting context about your codebase. Coval gives them that — and gives you proof.
Stops hallucinations before they land
Flags imports, methods, and environment variables the AI invented that don't actually exist in your project.
Keeps your AI's context true
Maintains a grounded project briefing from real analysis, so your agent stays oriented as the code changes instead of guessing.
Keeps your code AI-legible
Surfaces duplication, dead code, and architecture drift before they rot the codebase and confuse the next agent.
Gives you one number to trust
A single health grade with a trend over time, so you know whether AI-assisted changes are improving the code or eroding it.
A real analysis engine, not regex
Genuine graph + spectral analysis of your actual code — ranked, framework-aware, and tuned to your conventions.
Hallucination detection
Catches AI-invented APIs — imports, methods, named exports, and environment variables that don't exist.
Duplicates
Finds duplicate and near-duplicate code the AI re-created instead of reusing, grouped and ranked so you fix the real ones first.
Dead code
Surfaces unused and unreachable functions — and knows your framework's entry points (Next.js, Remix, SvelteKit, NestJS, and more) so it won't flag the wrong things.
Architecture drift
Highlights tangled structure, overloaded "hub" functions, misplaced code, and tightly coupled files — the drift that makes a codebase harder for both people and agents.
Dependency & tooling health
Unused and undeclared dependencies, project hygiene checks (lint, formatting, strict typing, CI), and an opt-in outdated-dependency scan.
Health grade
One 0–100 score and letter grade, tracked over time as a trend you can act on.
Smart noise control
Learns your project's conventions and intent — and recognizes common, intentional patterns — so it suppresses false positives instead of nagging you.
One engine, everywhere you code
The same local analysis renders in your editor, your terminal, and any agent that speaks MCP.
MCP server
coval-mcpPlug Coval straight into Claude Code, Cursor and Codex so your agent works against grounded context. Live analysis tools today; grounded-context generation and change self-checks next.
Embedded agent
A Claude-powered assistant that can scan, preview a fix, apply it, and open a fix branch or PR for you.
VS Code extension
Findings in the gutter, hover & inline insights, a live health report, impact analysis, and an architecture map — right in your editor.
CLI
covalA terminal scan with a clean, readable report. Drops into CI to gate on serious findings before they merge.
Web dashboard
Cloud-synced reports, history, accounts, and team plans.
JS/TS first, more in progress
JavaScript / TypeScript (including JSX / TSX)
Python · Go · Rust · Java
Structure & architecture insights. Some checks — like hallucination detection — are JS/TS-only for now.
Free, and built to be used
Coval's core — the CLI, MCP server and VS Code extension — is free, with no account required. It plugs into the AI tools you already use, and you decide what, if anything, ever syncs to the cloud.
Gate the merge, not just the editor
The CLI emits the same report your editor shows. Drop it into CI to catch hallucinated APIs and serious findings before AI-assisted changes ever merge.
# interactive scan npx coval scan . # gate CI on serious findings npx coval scan . --ci ✗ 2 hallucinated imports · build failed ✓ Grade B+ · trending up