Devotd Github
NewAI

anisayari/sparky 1.7.0 — Changes to temporary chat and RAG behavior, and a release smoke test

The personal AI assistant anisayari/sparky, which runs locally on Windows, changed the behavior of temporary chat and document indexing in 1.7.0 and added a release smoke test.

Repositoryanisayari/sparky ↗

The personal AI assistant anisayari/sparky, which runs locally on Windows, changed the behavior of temporary chat and document indexing in 1.7.0 and added a release smoke test.

GitHub description: 🔒 Your private AI assistant, 100% local on Windows: chat, vision, documents (RAG), web search, images, videos with sound, voice, code projects, MCP tools — one-line install, models picked for your PC.

stars
★ 42
Forks
1
Language
Python
License
MIT
Read original source ↗View on GitHub ↗

anisayari/sparky is a personal AI assistant that runs only locally on Windows PCs. PR #2, merged on November 3, 2026 UTC, is the Sparky 1.7.0 release, and this commit contained 6464 lines added and 1215 lines deleted across 30 files.

The user-facing changes in 1.7.0 described in the README concern conversations. Temporary chat uses its own temporary files instead of 'never being saved', and handles only text and images. Image generation and editing must be performed in normal chat, because the shared ComfyUI engine leaves its inputs and outputs behind. You can branch off from a message into a separate conversation, generated files follow the selected version, and Stop interrupts the turn in progress. Document search (RAG) indexes local folders connected in Workspace > Documents, covering additions, modifications and deletions, and upgrading an existing library requires reselecting the collection for each conversation that needs it.

The multi-file project checker (app/checker.py) reduces the way it used to actually run projects, checking only source syntax and local resource references, and returns errors to the assistant so that it can request repairs. Native execution and package installation are turned off in the checker. New files include app/chat_state.py, app/rag_sync.py, app/turn_runtime.py and app/process_owner.py. CI (.github/workflows/ci.yml) was changed to run on v* tag pushes as well, and Windows Python 3.11 was added to the matrix. A preview integration test that uses the Edge browser, a step that downloads the Piper voice fixture (en_US-lessac-medium.onnx) from the pinned commit e7229f7d and verifies its SHA-256, and a release-smoke job (with a 20-minute limit) that installs a Windows ZIP built with git archive, confirms that llama-server.exe is present and that ComfyUI is absent, and then checks the local port 7860 response were added.

However, the smoke test applies the chat-only installation (-ChatOnly -NoModels) to HEAD's source archive, so the Studio-side ComfyUI installation and model download path are not verified in CI. Multi-file projects also no longer run, so whether the result actually works cannot be known from this commit alone. The recorded information contains no license, star count or performance measurements.