Parameters and stop tokens
Parameters and stop tokens do not work when using a custom backend (llama.cpp on my home computer).
I could not find the option to add either of the aforesaid at all when running a local model.
Please fix.
I could not find the option to add either of the aforesaid at all when running a local model.
Please fix.
Ai
Liquid Apollo offers local ai models for iPhones. I tried 1.5b parameters models and they worked well, but I don’t recommend for code because of hallucinations.
Great
Amazing
Amazing
Incredibly powerful app. It truly hits different to talk to an AI running on your smartphone when you’re out in the middle of nowhere with no internet connection whatsoever. The models are very impressive with their capabilities for the small parameter count.
REQUEST: please allow us to switch the default model. It resets after every new chat! When using this with a custom backend it takes like 4 clicks each chat just to switch it, which makes an otherwise great feature more annoying to use.
REQUEST: please allow us to switch the default model. It resets after every new chat! When using this with a custom backend it takes like 4 clicks each chat just to switch it, which makes an otherwise great feature more annoying to use.
Show more
Brilliant Concept (I have some suggestions)
Honestly, this app utilizes “Apple Intelligence” (or, Apple’s integrated support for local AI via certain hardware and software additions and changes) far better than any other application. It literally allows you to use supported models locally, on device, free of charge. Apple themselves is yet to even improve Siri and allow their own very flawed voice assistant to make use of this. I paid for the new iPhone assuming this would be utilized, and so far, this app is probably the only application which makes ample use of that.
Suggestions:
- Enable sideloading of supported models by means of HuggingFace access (gguf, safetensor, etc… whatever is supported)
- Addition of more support for inference or API based model providers (aside from OpenRouter, perhaps Kimi/Z.AI/OpenCode’s plans which offer more reasonably priced access, Venice AI, etc)
- Addition of more tools (such as how the web search tool works; adding more would be nice)
- Enable MCP support (the search tool is cool, but not good enough — enabling the use of MCPs could help; perhaps add an addon store)
- Enable scripting for custom agents, tooling, etc (I’d prefer an “OpenCode”-esq agentic experience over the current agent, so customization of behavior, guiding reasoning, allowing custom tooling using premade SDK and otherwise, etc would be very cool)
I was thinking about making an open source version of this app and integrating such features, but clearly the developers of this application are competent and can do it themselves as they wish.
Suggestions:
- Enable sideloading of supported models by means of HuggingFace access (gguf, safetensor, etc… whatever is supported)
- Addition of more support for inference or API based model providers (aside from OpenRouter, perhaps Kimi/Z.AI/OpenCode’s plans which offer more reasonably priced access, Venice AI, etc)
- Addition of more tools (such as how the web search tool works; adding more would be nice)
- Enable MCP support (the search tool is cool, but not good enough — enabling the use of MCPs could help; perhaps add an addon store)
- Enable scripting for custom agents, tooling, etc (I’d prefer an “OpenCode”-esq agentic experience over the current agent, so customization of behavior, guiding reasoning, allowing custom tooling using premade SDK and otherwise, etc would be very cool)
I was thinking about making an open source version of this app and integrating such features, but clearly the developers of this application are competent and can do it themselves as they wish.
Show more
