Clairvoyance v0.85.0

Local AI hosting issues

from Stardock Forums

Question regarding hosting models locally and fine-tuning the model settings for increased performance within clairvoyance:

When hosting local models and connecting them to Clairvoyance I am artificially limited at a small token count when using llama.cpp and lmstudio server.  the models vary from 8,000 context tokens to 6,000 context tokens depending on if I load Gemma, MedGemma or Qwen3.8.  My machine has an RTX 5090 so I know it wasn't that I was hitting any limit due to hardware.  

I see no way to manually adjust the context token allocation within the GUI.  

I pass the token limit from the host software but Clairvoyance overrides this and sets an artificial limit on context. 

Questions:
1. Is there a way to set custom context tokens to match the amount specified in the local model instructionset?

2. Is there a granular GUI panel that allows for more control over the models we bring in locally?

3. Is there a hidden JSON file that allows for manual configuration of the local model properties?

4. Is there a way to disable reasoning of the self hosted Qwen3.8 model via ollama? (thinking chews through tokens)

Moved to Clairvoyance area

Stardock Software Type: Question Answered
324 views 4 replies
Reply #3 from Stardock Forums Top
Answer



1. Not yet. Custom context-token allocation is not exposed in this release. The **Capabilities** selector does not change the model’s context window.

2. No granular panel yet. Settings → Local AI handles discovery, installation, endpoints, and activation. Staff settings cover model, capabilities, and permissions—not temperature, context, output limits, or reasoning.

3. No supported hidden JSON file for manually configuring those model properties in this release. Ollama models can be customized externally with an Ollama `Modelfile`, but Clairvoyance has no supported property file for this.

4. Ollama can disable Qwen 3 thinking, but Clairvoyance currently has no switch for it. Direct Ollama usage supports `--think=false`, `/set nothink`, or API `"think": false`. Ollama documentation -https://docs.ollama.com/capabilities/thinking

+1 Loading…
Jump to answer in thread
Reply #1 from Stardock Forums Top

Hello,

Sorry to hear you are having issues. I have forwarded your problem/question to Stardock Support Team for their assistance. Please keep an eye on this thread for any updates. We appreciate your feedback and patience. Thank you.

Basj,

Stardock Community  Assistant

+1 Loading…
Object Desktop - Get all of our enhancement utilities in one suite
Reply #3 from Stardock Forums Top
Answer



1. Not yet. Custom context-token allocation is not exposed in this release. The **Capabilities** selector does not change the model’s context window.

2. No granular panel yet. Settings → Local AI handles discovery, installation, endpoints, and activation. Staff settings cover model, capabilities, and permissions—not temperature, context, output limits, or reasoning.

3. No supported hidden JSON file for manually configuring those model properties in this release. Ollama models can be customized externally with an Ollama `Modelfile`, but Clairvoyance has no supported property file for this.

4. Ollama can disable Qwen 3 thinking, but Clairvoyance currently has no switch for it. Direct Ollama usage supports `--think=false`, `/set nothink`, or API `"think": false`. Ollama documentation -https://docs.ollama.com/capabilities/thinking

+1 Loading…