E.V. v1.0.0 — first stable release

E.V. is a Windows-only, local-first voice assistant — Peter Parker's homemade AI, running on your own PC. Local Whisper speech recognition, a local (Ollama) or Claude brain, ElevenLabs or browser voice, deterministic PC control, long-term memory, and an Electron cyber HUD. Free and offline by default — no keys, nothing leaving your machine.

This is the first stable, tagged release. It hardens the app and makes the brain's quality measurable.

🔒 Security

• WebSocket handshakes from a foreign origin are rejected, so a random page in your browser can't drive the assistant or bypass the spoken confirmation.
• Ambiguous replies to a destructive action now cancel instead of running — only a clean "yes" proceeds.
• Apps launch only from the known list; the web-content summarizer treats page text as untrusted (prompt-injection hardening).
• Electron grants only the microphone permission.

⚙️ Reliability

• Greetings no longer freeze the server while weather loads.
• Fixed a Playwright page/memory leak on search and news.
• Conversation history is trimmed to fit the context window (no more silently forgetting the start of a chat).
• memory.json is written atomically.

🧭 Command router

• "don't forget X" now remembers instead of deleting.
• "restart spotify" restarts the app, not your PC; "play a game" no longer hits the media key.

🖥️ UI

• Dashboard stats regrouped so system metrics and weather are no longer mixed in one row. Wind now reads km/h.

🧪 Measurable brain

• A reproducible action-tag benchmark (eval/) with a results table in the README — and a self-serve command so you can benchmark your own card and model, not just trust the numbers.
• A pytest suite (43 cases) for the command router.

---

Free and local by default. Optional upgrades: a Claude key (screen vision / smarter brain) and an ElevenLabs key (nicer voice). See the README to get started, and CHANGELOG.md for the full list.

Stars, issues, and forks welcome. 🕸️