Private by default. Useful by design.
Local-first code audits, deliberate AI context, reviewable changes, and release checks.
Website · Documentation · Releases · Security
Localis starts without a model. It scans a repository locally, reports deterministic evidence, and makes every network boundary explicit. Connect Ollama, LM Studio, OpenAI, OpenRouter, or another OpenAI-compatible API only when a task needs a model.
| Workflow | What you get |
|---|---|
localis audit |
Local findings with file, line, rule, severity, and remediation |
localis privacy |
Selected files, redaction counts, endpoint, and payload SHA-256 |
localis ask |
Project-aware answers using only approved context |
localis propose / fix |
Schema-constrained change plans that cannot write by themselves |
localis apply / undo |
Diff preview, hash checks, private backups, and conflict-safe restore |
localis test / ship |
Test discovery and one release-readiness decision |
These captures come from Localis 0.2.0, not a product mockup.
Localis requires Node.js 20.9 or newer.
npm install --global localis
localis doctorbun add --global localis
localis doctornpx localis audit .curl -fsSL https://raw.githubusercontent.com/KebiLab/Localis/main/scripts/install.sh | shirm https://raw.githubusercontent.com/KebiLab/Localis/main/scripts/install.ps1 | iexDownload the Windows, macOS, or Linux installer from GitHub Releases. The first public builds are unsigned previews, so the operating system may show a warning.
After the first Windows installer is accepted into the WinGet community repository, installation will be:
winget install KebiLab.Localiscd your-repository
localis audit .
localis privacy . --file src --file package.json
localis ship .Machine-readable output is available for automation:
localis audit . --json
localis ship . --jsonOllama is the default local provider. LM Studio is restricted to loopback endpoints as well.
localis models
localis ask "Explain the authentication flow" . --file src/authFor an OpenAI-compatible API, keep the key in an environment variable:
$env:OPENAI_API_KEY = "your-key"
localis models --provider openai-compatible `
--endpoint https://api.openai.com/v1 `
--api-key-env OPENAI_API_KEYUse --dry-run to inspect the exact manifest and payload hash without contacting a provider.
- Deterministic audits do not need a model or network connection.
- Remote provider URLs must use HTTPS; local providers must use loopback.
- API keys are read from an environment variable or held in desktop memory for the current session.
- AI output becomes a validated plan, never an implicit file write.
applypreviews the diff, verifies original SHA-256 values, and requires confirmation.- Backups live under the ignored
.localis/directory andundorefuses unsafe overwrites.
See Privacy, Change plans, and Security policy.
apps/
cli/ published as `localis`
desktop/ Tauri 2 + React desktop application
web/ Next.js website and documentation
packages/
core/ published as `localis-core`
docs/ architecture, privacy, release, and change-plan details
scripts/ curl, PowerShell, and WinGet release helpers
npm install
npm test
npm run typecheck
npm run buildStart the website with npm run dev:web. Native desktop development additionally needs Rust and the Tauri prerequisites, then runs with npm run dev:desktop.
Localis 0.2.0 is a public preview. The deterministic scanner, privacy gateway, provider discovery, change plans, apply/undo, test discovery, ship gate, desktop app, and CLI are implemented. Code signing, WinGet approval, and editor integrations follow the first public installer feedback.
Read CONTRIBUTING.md before opening a pull request. Security reports belong in the private process described in SECURITY.md.
Made by KebiLab.
Localis was inspired in part by the open-source work and ideas shared by Andrew-py-dew. Localis is an independent implementation and does not include that repository's source code.
Licensed under the Apache License 2.0. Copyright © 2026 KebiLab.