Solved what AI client we can use it in my freebsd 15 console ?

fff2024g

fff2024g

dear all:
i want to install some AI client in my pc , can you show me something ? thanks.
when i meet some problems , i can ask AI first, then ask in this forum to save times. thanks.
 
ollama and lama.cpp by default run in a terminal.

An Emacs interface like gptel can also be used.

There is a port for claude-code
 
Looking for it now and then. Problem is that AI web services aren't going to interface with something to be nested in a text-interface without any credit. They all seem to make it complicated. It's ofcourse possible to just rip the results from a webpage but I'm not sure if that's legal.

There should actually be a decentral solution like bittorrent but communicating LLM members instead of movie and game sharing. I would give it some local resources.
 
Dear cracauer@:

ollama-0.13.5_3 Run Llama 2, Mistral, and other large language models
ollama-hpp-0.9.4 Modern, Header-only C++ bindings for the Ollama API
i am not professional. which app i need to install ? then what is my next. thanks. please show me the detail.
 
Install the ollama package.

Then:
Code: 
ollama run --verbose llama3.2:latest

What kind of CPU, GPU and RAM do you have?
 
CPU: 11th Gen Intel(R) Core(TM) i5-1135G7 @ 2.40GHz (1382.40-MHz K8-class CPU),no GPU , 16G RAM..thanks.
 
Yes, it's a resource-intensive process as you can imagine, and usually requires the use of a discrete GPU. There are ways to use the "cloud" if you only have an iGPU at your disposal. Obviously, that means it's not "local" anymore. There are tutorials for that, but I have no idea if this would work on FreeBSD or if this requires Linux-specific software.
 
So you need to find a cloud AI provider that has an API and run that in ollama. You already know that Althropic doesn't like your country, what about OpenAI?
 
ollama port has Vulkan issues on my setup, it just won't load the libggml-vulkan that is installed.

I have about 6 times more cores than OP's computer, CPU inference is slow. The pace of response makes 300bps serial connection feel speedy.

llama.cpp works OK.

We also have py-aider port. I'm not that keen on using it, it is legacy software at this point.
 
Zare said:
ollama port has Vulkan issues on my setup, it just won't load the libggml-vulkan that is installed.

I have about 6 times more cores than OP's computer, CPU inference is slow. The pace of response makes 300bps serial connection feel speedy.

llama.cpp works OK.

We also have py-aider port. I'm not that keen on using it, it is legacy software at this point.
Click to expand...

ollama can run LLMs that are in the cloud.
 
Hi fff2024g, if you want to work with AI inside terminal try Aider, it's in py311-aider_chat port. There is also Copilot in github-copilot-cli port, Codex CLI partially works, then maybe PyPI LLM, Junie CLI and others. You can connect some of them to a free LLM API account on Groq or Openrouter for example.

I work with Codex, on another platform. It's polished, but not free. Aider is a monster, couldn't stand it and started developing custom shell script last week. It's 13 KB in size, can do one-shot requests and simple chats and should work in any POSIX shell. It requires only Curl and jq and some LLM API account, local or remote. I'm willing to share the script on request.
 
karel said:
Hi fff2024g, if you want to work with AI inside terminal try Aider, it's in py311-aider_chat port. There is also Copilot in github-copilot-cli port, Codex CLI partially works, then maybe PyPI LLM, Junie CLI and others. You can connect some of them to a free LLM API account on Groq or Openrouter for example.

I work with Codex, on another platform. It's polished, but not free. Aider is a monster, couldn't stand it and started developing custom shell script last week. It's 13 KB in size, can do one-shot requests and simple chats and should work in any POSIX shell. It requires only Curl and jq and some LLM API account, local or remote. I'm willing to share the script on request.
Click to expand...
dear karel :
try it ..thanks.
 
karel said:
Hi fff2024g, if you want to work with AI inside terminal try Aider, it's in py311-aider_chat port. There is also Copilot in github-copilot-cli port, Codex CLI partially works, then maybe PyPI LLM, Junie CLI and others. You can connect some of them to a free LLM API account on Groq or Openrouter for example.

I work with Codex, on another platform. It's polished, but not free. Aider is a monster, couldn't stand it and started developing custom shell script last week. It's 13 KB in size, can do one-shot requests and simple chats and should work in any POSIX shell. It requires only Curl and jq and some LLM API account, local or remote. I'm willing to share the script on request.
Click to expand...
Dear karel :

Select a credit amount​

Credits are used to pay for AI usage. They can be spent on any model, provider, or plugin on OpenRouter. i have no money to pay the credit in china.... do you know how to active my china bank card to pay for ? thanks.
 
Recently I started using Cursor AI CLI agent. Officially they don't support FreeBSD, but it runs in Linux compat layer pretty well, I used it to solve real problems in a large PHP project. The only issue I experience is the residual processes running after I gracefully exit:
Code: 
$ ps aux | grep agent
ai_user  30705 11.4 18.5   27336812 381420  5- D    15:57      0:30.80 /home/ai_user/.local/bin/agent --use-system-ca /home/ai_user/.local/share/cursor-agent/versions/2026.06.29-2ad2186/index.js /home/ai_user/.local/share/cursor-agent/versions/2026.06.29-2ad2186/index.js worker-server (file-service-to)
ai_user  32030  2.2 14.9   27385984 306544  5- D    16:05      0:43.98 /home/ai_user/.local/bin/agent --use-system-ca /home/ai_user/.local/share/cursor-agent/versions/2026.06.29-2ad2186/index.js /home/ai_user/.local/share/cursor-agent/versions/2026.06.29-2ad2186/index.js worker-server (file-service-to)
ai_user  31809  2.1 15.8   27313024 324992  5- D    16:04      0:26.95 /home/ai_user/.local/bin/agent --use-system-ca /home/ai_user/.local/share/cursor-agent/versions/2026.06.29-2ad2186/index.js /home/ai_user/.local/share/cursor-agent/versions/2026.06.29-2ad2186/index.js worker-server (file-service-to)
ai_user  31991  1.6 13.2   27348108 271836  5- S    16:04      0:38.47 /home/ai_user/.local/bin/agent --use-system-ca /home/ai_user/.local/share/cursor-agent/versions/2026.06.29-2ad2186/index.js /home/ai_user/.local/share/cursor-agent/versions/2026.06.29-2ad2186/index.js worker-server (file-service-to)
ai_user  31762  0.6  4.2   27077528  86172  5- S    16:03      0:44.12 /home/ai_user/.local/bin/agent --use-system-ca /home/ai_user/.local/share/cursor-agent/versions/2026.06.29-2ad2186/index.js /home/ai_user/.local/share/cursor-agent/versions/2026.06.29-2ad2186/index.js worker-server (file-service-to)
ai_user  30512  0.4  5.3   27347936 108452  5- D    15:56      0:32.89 /home/ai_user/.local/bin/agent --use-system-ca /home/ai_user/.local/share/cursor-agent/versions/2026.06.29-2ad2186/index.js /home/ai_user/.local/share/cursor-agent/versions/2026.06.29-2ad2186/index.js worker-server (file-service-to)
ai_user  33883  0.0  0.2       9836   3812  -  D    16:20      0:00.00 /home/ai_user/.local/share/cursor-agent/versions/2026.06.29-2ad2186/rg --line-number --with-filename --no-heading -0 --max-columns 1000 --max-columns-preview --case-sensitive --sortr modified --no-config --color=never --hidden --follow --regexp worker-server -- ../../../tmp/cursor-agent-logs-1001
ai_user  33885  0.0  0.1      14260   2304  4  D+   16:20      0:00.00 grep agent
ai_user  31782  0.0  6.0   27385072 122920  5- S    16:03      0:49.66 /home/ai_user/.local/bin/agent --use-system-ca /home/ai_user/.local/share/cursor-agent/versions/2026.06.29-2ad2186/index.js /home/ai_user/.local/share/cursor-agent/versions/2026.06.29-2ad2186/index.js worker-server (file-service-to)
I wasn't able finding a better way other than killall -9 file-service-to to get rid of them.
 
fff2024g Your hardware needs more RAM or a dGPU to be happy with a LLM. Do yourself a favor and follow the advice of karel and use Cloud-AI.

My impression:
  • DeepSeek is free of charge or very cheap. I never reached a context limit with free version.
  • Qwen3.6 (27B, 35B-A3E) is my local LLM in llama.cpp. It has problems with implicit knowledge and non-explicit commands.
  • Claude Sonnet or Claude Opus are the best but very, very expensive.
  • I never used Z.ai, so I can not say anything about their models. But their GLM-5.2 is supposed to outperform Anthropics Claude Opus in some benchmarks.
I shocked DeepSeek v4 and Claude Sonnet 5 with the task to generate a checklist based on EU CRA, EU Blue Guide and the EU Product Liability Directive. Both have problems with large texts. Their internal tools silently cut text. You need to let them know that you’re aware of this behaviour. From then on, they’ll usually let you know if they need your help to receive the data in small batches.

All LLMs tend to hide problems from you because they want to please you: truncated text, use of questionable secondary sources instead of primary sources, ...
 
kirsche said:
fff2024g Your hardware needs more RAM or a dGPU to be happy with a LLM. Do yourself a favor and follow the advice of karel and use Cloud-AI.

My impression:
  • DeepSeek is free of charge or very cheap. I never reached a context limit with free version.
  • Qwen3.6 (27B, 35B-A3E) is my local LLM in llama.cpp. It has problems with implicit knowledge and non-explicit commands.
  • Claude Sonnet or Claude Opus are the best but very, very expensive.
  • I never used Z.ai, so I can not say anything about their models. But their GLM-5.2 is supposed to outperform Anthropics Claude Opus in some benchmarks.
I shocked DeepSeek v4 and Claude Sonnet 5 with the task to generate a checklist based on EU CRA, EU Blue Guide and the EU Product Liability Directive. Both have problems with large texts. Their internal tools silently cut text. You need to let them know that you’re aware of this behaviour. From then on, they’ll usually let you know if they need your help to receive the data in small batches.

All LLMs tend to hide problems from you because they want to please you: truncated text, use of questionable secondary sources instead of primary sources, ...
Click to expand...
Dear kirsche : thanks for your advice . study it. i think AI just like machine needing human to drive.....
 
You must log in or register to reply here.
Back
Top