← Back to Advisor Prep Hero

Changelog

What's shipped in Advisor Prep Hero, version by version. The honest list, not just the highlights.

Last updated 2026-05-04 · For the full developer changelog, see CHANGELOG.md in the repo.

v2.0.0 May 2026 Latest

The big one. Multimodal chat, PDF chat with RAG, long-context Compress, read aloud, a plugin system, a templates marketplace, a web demo, and Spanish + German.

Added: Multimodal AI input

Added: PDF chat

Added: PDF RAG

Added: Long-context [Compress]

Added: Read aloud (TTS)

Added: Templates marketplace

Added: Plugin system

Added: Plugin marketplace

Added: Plugin developer ecosystem

Added: Try in browser

Added: Spanish + German

Added: Mobile access docs

The full announcement post: Advisor Prep Hero reads images, PDFs, and runs plugins now.

v1.7.2 April 28, 2026 Shipped

Lead capture + funnel telemetry. Opt-in only.

Added

Privacy stance

Telemetry default is OFF. Without explicit consent, no events are ever sent. Local-first promise stays intact.

v1.7.1 April 28, 2026 Shipped

Trial visibility: surface the countdown outside Settings.

Added

Changed

v1.7.0 April 27, 2026 Shipped

Pricing model rewrite + honest license-state UI.

Added

Changed

v1.6.0 April 27, 2026 Shipped

First commercially-launched release. Mac support, license activation, and 6+ rounds of polish.

Added

Tab + group UX overhaul

Onboarding polish

Fixed

v1.5.0 April 16, 2026 Shipped

The biggest Advisor Prep Hero release yet. Four headline capabilities + 18 quality-of-life extras + 2 multi-model mediums.

Four flags

1. Memory: the AI workspace that remembers your stuff

Local RAG over your notes via LanceDB + fastembed-rs (M1), an @workspace chat command + per-chat "Ask my workspace" toggle (M2), and a user-approved memory facts file always in the system prompt (M3). Embeddings run on-device with a 384-dim ONNX model (intfloat/multilingual-e5-small). Per-workspace vector store under .keepance/vectors/. Nothing sent off-device.

2. MCP server: your workspace, available in every AI tool you use

A real JSON-RPC 2.0 MCP server binary exposing five tools (list / read / search / write-with-approval / facts) over stdio. Shipped as a per-platform .mcpb Desktop Extension bundle for Claude Desktop, Cursor, Zed, and any other MCP-compatible client. One-click install from Settings → Integrations.

3. Side-by-side AI editing

Select text, ask for a revision, watch the streaming diff land where your cursor was, accept or reject each hunk individually. Every accepted hunk is written to version history with author: 'ai', the original prompt, the provider + model, and the offset range. Full audit trail. Hotkey: Ctrl+Shift+E.

4. Voice input + local models (Ollama)

Press-to-talk voice capture via bundled Parakeet.cpp / whisper.cpp sidecar. Audio bytes never leave the machine. Ctrl+Shift+Space to record + insert; Ctrl+Shift+N to save transcript to Inbox/. Plus Ollama as a first-class fourth provider that auto-detects models from your local install, $0 per call, fully offline.

Eighteen Quick Wins

Two multi-model mediums

v1.0.8 April 16, 2026 Shipped

Auto-updater + document suite + branded start screen + settings system.

Added

v1.0 → v1.0.7 April 2026 Shipped

First commercial pipeline + cross-platform code signing.