Mori

Mori finds source fragments with similar structure—even when they use different names or programming languages—and gives you an explainable shortlist to review.

Mori reports structural evidence. A match is never proof that two fragments behave the same way or should be merged.

Install

Download a native archive from the latest release, or build from source with Go 1.23+ and a C compiler:

go install github.com/Cyberlane/mori/cmd/mori@latest
mori version

Official release binaries include complete build provenance and are recommended for CI and auditable reports.

Each release also includes checksum-pinned Homebrew, Scoop, and WinGet manifest assets, an SPDX SBOM, and GitHub/Sigstore attestations. Package-index submission is intentionally separate, so verify the release asset before local use.

Debugging and support

Mori v0.33.0 can capture an optional, source-free diagnostic session for a scan. Put --diagnostics before other scan options so argument errors can be captured:

mori scan --diagnostics session.json --profile review .
mori support bundle --session session.json --output mori-support.zip
mori support inspect mori-support.zip

Inspect the bundle before attaching it to a support conversation. Nothing is uploaded automatically. See support reports for the exact contents, limits, and optional reviewed-finding labels.

Quick start

From a project root, inventory the source and preview a project setup:

mori inspect .
mori setup

mori setup inventories the project, asks a few focused questions, previews a conservative .mori.json, and writes it only after confirmation. Review its exclusions after the first scan; Mori cannot decide which tests, generated files, migrations, or framework repetition are intentional in your project.

For a library or monorepository, choose the suggested library scope when its source roots match your intent. Choose application for broader non-test code, including demos and tooling. Review the proposed roots and exclusions, then run the exact command printed by setup. For an application scope, retain a first review:

mori scan --scope application --format agent --output review.json

Keep review.json local and inspect its coverage as well as both candidate locations. If you kept the inclusive configuration, omit --scope application. For very large layouts, choose a smaller source root before comparison. Library suggestions and future-proof test patterns require Mori v0.33.0 or later. See the published documentation and focused first-review guide.

For a no-write trial:

mori scan --profile review .

The review profile starts with same-language code, an 85% threshold, a 40-token floor, generated-source exclusion, and required aggregate coverage. It still includes tests and stories. In a large repository, start with a reviewed source root such as mori scan --profile review packages/core/src; inspect both sides of the first groups before deciding which categories to exclude. See the first-review guide for scope and staged-review tradeoffs.

For a cross-language review, choose it explicitly:

mori scan --threshold 0.70 --cross-language-only examples/email-validation

Illustrative cross-language result (scores depend on source and version):

1. 92.7% structural similarity · 1 location pair(s)
   weighted feature evidence: 114 intersection / 123 union
   A  Validator.java:6-9   [java]       looksLikeEmail
   B  validator.js:1-4     [javascript] looksLikeEmail
   shared shape: 3 calls, 1 return, 2 bindings

Always read both locations. Mori does not establish equivalent runtime values, effects, types, external calls, permissions, transactions, or error behavior.

Choose a workflow

Lower the token floor toward 12 only for deliberate broad exploration; small callbacks and wrappers commonly dominate at that size.

Use mori explain <content-pair-id> [scan options] . to reproduce and isolate one reported identity. Text output uses restrained color on terminals; set --color never or NO_COLOR to disable it.

Supported source

Mori currently supports:

• Bash/POSIX shell and Zsh;
• C, C++, C#, Dart, GDScript, Go, Hack, Java, JavaScript/JSX, Kotlin, Lua, Luau, PHP, PowerShell, Python, Ruby, Rust, Swift, TypeScript/TSX;
• generic SQL queries and explicitly selected PostgreSQL queries.

Run mori languages for the exact extensions, families, fragment kinds, and extensionless shebangs supported by your installed version. See Languages and parser limits for comparison boundaries and known gaps.

Editors and AI coding tools

The repository includes a dependency-free VS Code reference client. It analyzes unsaved buffers through a local Mori process and reports SARIF diagnostics without uploading source.

Install Mori's review skill into a project for compatible coding agents:

mori skill install --project .

The skill teaches agents to treat matches as review leads, verify coverage, inspect both source locations, and avoid score-only refactors.

After installing a newer Mori binary, coordinate the project pin, embedded skill, configuration and baseline validation, and automation inventory with:

mori project upgrade --dry-run .
mori project upgrade --apply .

Apply mode updates .mori-version, the tracked .mori-project.json contract, and only a missing or recognized Mori-managed Agent Skill, with backups. Unknown skill changes fail closed. It does not install the CLI, invent configuration policy, rewrite hooks or CI, commit, push, or release anything.

To let a project agent configure Mori without granting hidden write access:

mori setup --agent --format json .

The agent can answer the emitted questions and preview the exact configuration before an explicit --apply. See Editors and coding agents.

Documentation

• Documentation home
• Getting started
• Reviewing results
• SQL and embedded SQL
• Parser compatibility
• Automation and baselines
• Editors and coding agents
• Project contract and upgrades
• Project configuration
• Scoring
• Machine integration
• Architecture
• Adding a language

Smooth project reviews

Start a new integration with advisory staged review to see structural leads without blocking on similarity alone. Existing hooks remain strict. Both policies retain configured coverage requirements; the report separates policy success from analysis completeness. Trust the project version pin and schedule upgrades separately. Upgrade diagnostics identify obsolete workflow instructions without rewriting project policy.

Rollout guidance covers native verification, reviewed project migration and opt-in pilot evaluation.

Local feedback is off by default. Explicit per-project consent enables bounded measurements; export previews a separate minimized bundle. Mori does not upload feedback. Collection never accepts findings or changes a baseline. mori feedback summarize compares explicitly exported bundles offline without counting exports as distinct projects.

Development

make check

See Contributing for development policy. The calibration corpus is regression evidence for its reviewed cases, not a universal accuracy claim.

License

MIT