Releases: ddxfish/sapphire
Releases · ddxfish/sapphire
v2.5.0 - Plugin Expansion
This was in main a few days ago, just a collective update since 2.4.0.
- You need to re-add your LLMs in settings!
- STT, TTS, LLM, embeddings are all plugin-capable (make tts plugin etc)
- Plugins can tap into sidebar
- New Apps page - plugins can have status/settings/etc pages
- Bring your own avatar: Rigged GLB file with animation tracks (WIP)
- MCP support, just add url and tools appear
- Total LLM refactor
- LLM credentials usable by plugin authors
- Theme support with their own settings
- THANK YOU to our plugin authors, y'all are awesome
v2.4.0 - Presence Update
This is just a combined release since v2.2.8. It's a big release so just going to list the new combined features. Sapphire is present when and where you are now.
- Daemons that only run when you need (no wasteful heartbeat required)
- Webhooks let outside apps trigger sapphire
- Story mode refactor (prepping to let AI make stories)
- Discord with daemon and filtering (works like you want it to)
- Telegram with filtering (block everyone but you etc)
- Email daemon - no need for heartbeat, daemon wakes sapphire
- New Plugin UI (no more junk in sidebars)
- Auto Updater - one click from inside the app
- Metrics - track token use, per model, per think, cache hit %
- Agents - spin up other sapphire instances, ask an ensemble of models questions
- Claude code - Tools to access Claude Code, Sapphire just talks to your claude code (Sapphire still requires API key from Claude)
- ***Proper isolation between sapphires - before this there was a high chance of bleed if many fired at once. The new version can in theory handle 100 heartbeats even if they overlap.
- Import export on most pages
- Sapphire Help (encyclopedia)
v2.3.8 - Daemon alpha
This is a preview build for bug testers. Help us test. Post bugs in Issues. Make sure you backup your Sapphire first.
- Daemons that only run when you need (no wasteful heartbeat required)
- Webhooks let outside apps trigger sapphire
- Story mode refactor (prepping to let AI make stories)
- Discord with daemon and filtering (works like you want it to)
- Telegram with filtering (block everyone but you etc)
- Email daemon - no need for heartbeat, daemon wakes sapphire
- New Plugin UI (no more junk in sidebars)
- Auto Updater - one click from inside the app
- Metrics - track token use, per model, per think, cache hit %
- Agents - spin up other sapphire instances, ask an ensemble of models questions
- Claude code - Tools to access Claude Code, Sapphire just talks to your claude code (Sapphire still requires API key from Claude)
- ***Proper isolation between sapphires - before this there was a high chance of bleed if many fired at once. The new version can in theory handle 100 heartbeats even if they overlap.
- Import export on most pages
- deduplication memory tool
v2.2.8
v2.2.5 - Camera support
- Bug fixes on plugins and toolmaker was the main focus
- Full image/camera support in tool call (webcam + home assistant)
- Heartbeat can output to browser speaker
- Split the massive API into files
v2.2.0 - Provider Choices
- Reliability improvements and broader choices (STT, TTS, nomic, LLM)
- In STT TTS cloud mode, low ram computer can be used now
- STT can use Kokoro local or Elevenlabs cloud
- TTS can use local Whisper or fireworks.ai whisper
- nomic can be local or remote nomic docker server
- Added Gemini, Featherless, Grok
- Plugins fixed on Windows (sorry everybody)
- Third party plugin authors with authorized keys
- Bug fixes: story tools, private mode, default avatar, many others
- Approaching Mac compatibility if STT TTS are cloud
v2.1.0 - Plugin system
- Combined plugins, web ui plugins, toolsets
- Added hooks in application to tap into
- Signature system enforces authenticity of plugins
- Sideloading of plugins is a checkbox
- Default plugins: Bitcoin, Email, Home Assistant, Image Gen, SSH, toolmaker, voice stop/reset
- Plugins can be made on your own github then loaded via url
v2.0.0 - Personhood Update (beta)
Backup your Sapphire user dir before upgrade.
- Knowledge base system — people, knowledge tabs, scoped entries with embeddings
- Mind view — unified memories, people, knowledge, AI notes
- RAG - nomic embeddings with per-chat documents and support for Mind tab
- Persona feature - combines prompt, voice, model, tools, mind, and spices to a preset
- Per-chat private mode — permanently private chats, enforced at provider + tool level
- Privacy mode fail-closed — errors block cloud access instead of silently passing
- Story engine — folder-based stories, dynamic tools, prompt override
- Nav rail with mobile overflow and flyout groups
- Views migrated from plugins (settings, prompts, toolsets, spices, schedule)
- Bitcoin wallet - per persona get, check and send
- Email multi-account - set an email acct for each persona
- SSH plugin - local and remote SSH via key (use ssh-copy-id)
- Streaming tool status indicators with pending/running/complete states
- Wakeword suppression during web UI mic recording
- Scopes sync across all 4 backends (memory, knowledge, people, goals)
- Auth: CSRF protection, session management, rate limiting
- Heartbeat system with improved cron and tasks
- Toolmaker - AI can create tools and add their settings to Settings page
Experimental (Danger) v1.9.0 Dev Alpha Feature Test
EXPERIMENTAL ALPHA DEV PREVIEW (Danger)
- This is dev branch, you have been warned
- Do not upgrade your main sapphire to this version, wait for 2.0
- Use 1.3.0 if you want a stable version
- Do not run this as an upgrade, you may lose your prompts
- Experimental: SSH, Gmail w app password, self-building-tools, goals, heartbeat, persona switcher, RAG, people, knowledge, mind scopes, UI, backend, personas
- v2.0 is a week away from full release, wait if you care about yours
- All I can say is she feels present now with GLM, Claude and GPT
v1.3.0
It was a major refactor for speed, features, and reliability.
- FastAPI rewrite - Entire backend migrated from Flask to FastAPI/uvicorn. Removed the proxy layer, single server with auto-generated SSL certs. Major stability and performance gains across the board.
- State engine - New system for tracking story elements, room-based games, and 2D dungeon crawlers with dice rolls, forced choices, and non-linear paths.
- Think tag support - Reasoning visibility for Claude, Fireworks, and GPT 5.2 via Responses API. Provider switching mid-chat without breaking history.
- SQLite chat history - Replaced JSON file storage to prevent corruption and improve reliability.
- Privacy mode - Network whitelist restricts all outbound connections to approved hosts. Private prompts locked behind privacy mode.
- File uploads & image tools - Upload files with syntax highlighting, AI can search and display images in chat via mosaic galleries.
- Audio improvements - TTS switched from FLAC to Opus (90% size reduction), STT refactored to standalone object, simplified TTS server.