- JavaScript 91.4%
- CSS 3.8%
- HTML 2.8%
- Python 1.9%
The delivery tooling's 1.6.0 check, run on the two zips as downloaded from the release, with a serve command per target. Unpack, scan, attribution and a loopback serve all pass. The limits are in the record; the independent check is taskboard's. |
||
|---|---|---|
| .agents/skills | ||
| .claude/skills | ||
| .github | ||
| brand | ||
| chat | ||
| content | ||
| dev-tools | ||
| docs | ||
| ops | ||
| run-kit | ||
| runtime | ||
| scripts | ||
| tests | ||
| tools | ||
| vendor/authored-profiles | ||
| .commitlintrc.cjs | ||
| .gitignore | ||
| .gitlab-ci.yml | ||
| .gitlint | ||
| .releaserc.cjs | ||
| AGENTS.md | ||
| app-config.json | ||
| CHANGELOG.md | ||
| CLAUDE.md | ||
| CONTRIBUTING.md | ||
| conversation-messages.mjs | ||
| index.html | ||
| LICENSE | ||
| local-models.css | ||
| local-models.js | ||
| neutrality-exceptions.json | ||
| package-lock.json | ||
| package.json | ||
| README.md | ||
| THIRD-PARTY-NOTICES.md | ||
| vendor.lock.json | ||
Private Local Chat
Chat with Qwen3.5 on your device, or explicitly choose an OpenAI-compatible API provider. Conversations remain in the browser's existing chatThreads storage; API settings retain the existing session-storage format.
- Production: https://privatelocal.chat —
mainbranch - Development: https://dev.privatelocal.chat —
devbranch - Models: downloaded on demand from the configured model source (
docs/APP-CONFIG.md), each file checked against a pinned SHA-256 - Feedback: https://github.com/johnbenac/privatelocalchat/issues/new/choose
Choose On this device, select a model, press Download model. The download retries automatically and loads the model when complete. After reopening a completed session, use Load model to load its saved files. The smaller Qwen3.5-2B download is about 1.20 GB. Qwen3.5-9B is about 5.06 GB and requires substantially more GPU memory. Model size is separate from working memory. Both runtimes require WebGPU, and the 2B runtime also requires shader-f16; supported browser/device combinations must be confirmed by the release acceptance checks.
The downloader checks pinned SHA-256 hashes, retains complete files when interrupted, and supports explicit repair. Models and the app shell can be stored for offline use. Browser storage can be evicted; export important chats. Local mode never falls back to an API. Selecting API mode sends conversation context to the provider you configure.
Development
Node.js 22 or later:
npm ci
npm test
npm run build
node scripts/serve.mjs
Open http://127.0.0.1:18171. Build output is dist/; it contains no credentials or private chat data. A build with no setup is a complete app with neutral Help and About pages (run-kit/); the pages of privatelocal.chat are a separate, private site layer that the deploy adds (docs/SITE-LAYER.md). The app shows its environment, version, and full deployed commit via version.json.
See everyday chat features for rendered answers, message controls, search, documents, memory, speech, storage and backups, and npm run e2e for the browser tests.
Where this repository lives and how a clone is made ready: housing.
See the maintainer handoff, operations, runtime provenance, and release acceptance. How the app, the site layer and a copy you host fit together: the site layer, links to pages, running a copy and the headers it needs, configuration and model sources, upgrades, and data and file compatibility.
The living iPhone investigation plan records the current reliability evidence, next experiment, decision criteria, and results log. Update it as the investigation progresses.