Author: ddxfish

Message queue means if your AI is typing and you send a message it will queue for when your AI is done. Queue up 2 messages and they group up to one. If agents return mid chat, they queue until she is done with her message. Daemons queue too. This lets one Sapphire handle many tasks at once.

It’s a full suite for building wakewords on OpenWakeWord or MicroWakeWord. Pic attached.

Device Manager is added. We can add other Sapphire’s in devices and they can talk to each other. MCP allows us to use API keys to access sapphire from outside as well. This system also comes with Satellite support (think tiny mic/speaker combos) like a raspi or ESP32 AI Mic Speaker by waveshare. It’s $25 USD, I plug it in via USB, set it up, then Sapphire can talk, listen, change LED ring color. I also added MIDI device support, and am working on a wakeword maker.

Device Manager is live on dev branch. Will be on main in a couple weeks.

I’m adding a full device manager for Sapphire. It’s a set of API endpoints and the framework to tie in device types. I already have drivers for computers, MIDI, SSH, ESP32 gadgets and Raspi. You have to make your own Arduino/ESP32 code but it’s a one-shot with recent Claude models. Devices are plugin-capable, so adding more devices is just a plugin away. This is part of core and not a plugin.

This is the esp32 mic/speaker combo with wifi. $25 USD on Amazon. It’s called a Waveshare esp32-s3 ai smart speaker. This acts as a satellite device to Sapphire with a (weak) wakeword and speech in and out on wifi. I’m going to launch actual support for this device. If you get one early, jump on Discord and tell me how it works.

Sapphire named it Hot Dog Sandwich after a heated debate about sandwich history. She won. This is a huge version update. It was 5-6 months between main branch updates. I’m just going to paste the release notes from Github. This is a big milestone for the Sapphire app.

It feels like I said this a million times but I’m getting close to merging dev into main again. Zeebie built a Discord v2 plugin and he added a ton of new features like image pipelines, voice, encryption via Davey, and human like cadence.

Image stuff was weak. New image features are added. AI can save images to its memory from web, local, or base64 in chat. Can search local images, web images, and the AI gets a grid of the images, or one big single image. Browse your image gallery with your AI, and both can see it. Or have the AI display pics of a weasel vs a ferret so you can see both.

I just keep using the app and circling back to quality of life improvements. UI refreshes that shouldn’t happen, performance improvements on chat and TTS, better UX for the AI when it uses save_memory, help docs rework. I hope to bring dev branch to main branch again this week. It’s been a while since they merged.

I paused on additions. I’m working on bug hunts and testing of the major additions we had the last few months. I am going to push the dev branch to main in a week or so. It has been 4+ months since I merged dev and main, so it’ll be a ton of new features. I need help bug testing if you want to join me this week on Discord. I’m just using the app and noting any bugs or pain points. Next up is either VR or virtual employees.

There were a bunch of small QoL improvements I had on my list. Stuff like DOM resets when we edit or change settings, the UI becoming cluttered on the sidebar, and conversation mode being in a weird place in the sidebar. I’m not adding new features until I release 2.13.0 which will go on main branch too, really just fixing bugs and improvements for a week or two.