This is a dead project what is everyone using now?

[ColtonKinstley]

Not to take away from the usefulness of the project. It's still functioning fine in many ways. But, PRs aren't getting any love and there are some clear updates needed to support the gpt-5 class models.

[TheRob2D]

Hopefully it gets an update soon! Would love to see it get 5 support.

[JackTheTripperr]

There is a quick workaround available to use while waiting for the devs to get around to merging pull requests addressing this.

In app.py starting at line 36:

// app.py
...
    temperature: float = typer.Option(
-       0.0,        
+       cfg.get("DEFAULT_TEMPERATURE"),
        min=0.0,
        max=2.0,
        help="Randomness of generated output.",
    ),
...

Then, in config.py at line 26:

// config.py
...
DEFAULT_CONFIG = {
    # TODO: Refactor it to CHAT_STORAGE_PATH.
    "CHAT_CACHE_PATH": os.getenv("CHAT_CACHE_PATH", str(CHAT_CACHE_PATH)),
    "CACHE_PATH": os.getenv("CACHE_PATH", str(CACHE_PATH)),
    "CHAT_CACHE_LENGTH": int(os.getenv("CHAT_CACHE_LENGTH", "100")),
    "CACHE_LENGTH": int(os.getenv("CHAT_CACHE_LENGTH", "100")),
    "REQUEST_TIMEOUT": int(os.getenv("REQUEST_TIMEOUT", "60")),
    "DEFAULT_MODEL": os.getenv("DEFAULT_MODEL", "gpt-4o"),
+   "DEFAULT_TEMPERATURE": os.getenv("DEFAULT_TEMPERATURE", "0.0"),
    "DEFAULT_COLOR": os.getenv("DEFAULT_COLOR", "magenta"),
    "ROLE_STORAGE_PATH": os.getenv("ROLE_STORAGE_PATH", str(ROLE_STORAGE_PATH)),
    "DEFAULT_EXECUTE_SHELL_CMD": os.getenv("DEFAULT_EXECUTE_SHELL_CMD", "false"),
    "DISABLE_STREAMING": os.getenv("DISABLE_STREAMING", "false"),
    "CODE_THEME": os.getenv("CODE_THEME", "dracula"),
    "OPENAI_FUNCTIONS_PATH": os.getenv("OPENAI_FUNCTIONS_PATH", str(FUNCTIONS_PATH)),
    "OPENAI_USE_FUNCTIONS": os.getenv("OPENAI_USE_FUNCTIONS", "true"),
    "SHOW_FUNCTIONS_OUTPUT": os.getenv("SHOW_FUNCTIONS_OUTPUT", "false"),
    "API_BASE_URL": os.getenv("API_BASE_URL", "default"),
    "PRETTIFY_MARKDOWN": os.getenv("PRETTIFY_MARKDOWN", "true"),
    "USE_LITELLM": os.getenv("USE_LITELLM", "false"),
    "SHELL_INTERACTION": os.getenv("SHELL_INTERACTION ", "true"),
    "OS_NAME": os.getenv("OS_NAME", "auto"),
    "SHELL_NAME": os.getenv("SHELL_NAME", "auto"),
...
}

Finally, you'll want to edit your .sgptrc:

//.sgptrc
...
CHAT_CACHE_PATH=/tmp/chat_cache
CACHE_PATH=/tmp/cache
CHAT_CACHE_LENGTH=100
CACHE_LENGTH=100
REQUEST_TIMEOUT=60
DEFAULT_MODEL=gpt-5
+DEFAULT_TEMPERATURE=1.0
DEFAULT_COLOR=magenta
DEFAULT_EXECUTE_SHELL_CMD=false
...

[Sharon-Needles]

Deepseek_Coder 6.7b on llama.cpp. Arch Linux OS

…

On Tue, Sep 30, 2025 at 10:02 AM TheRob2D ***@***.***> wrote: Hopefully it gets an update soon! Would love to see it get 5 support. — Reply to this email directly, view it on GitHub <#725 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BLNA4WMYGNTYCKFZP2CQ6ED3VKEP7AVCNFSM6AAAAACH35BTB6VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTINJVGM2DAMA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[endolith]

I just use Open Interpreter to effectively replace the shell altogether (but that's also somewhat dead; trying to revive it because I don't know of anything else like it)

[Sharon-Needles]

Idk which OS you are using, so I will assume it’s a Debian. So I wrote three install one liners for each of the local model that I’m running. All uncensored and unfiltered. (I’m running Blackarch so if you want that one I will spin it up for you to save you some time.) ———— #DeepSeek Coder (one-liner for Kali/Debian): sudo bash -lc 'apt update && apt install -y git build-essential cmake python3 python3-pip rlwrap jq curl && [ -d /opt/llama.cpp ] || git clone https://github.com/ggerganov/llama.cpp.git /opt/llama.cpp && cd /opt/llama.cpp && cmake -S . -B build -DCMAKE_BUILD_TYPE=Release && cmake --build build -j"$(nproc)" && mkdir -p /opt/gguf_models /root/.cache/local_llm/history && touch /opt/gguf_models/deepseek-coder-6.7b-base.Q4_K_M.gguf && ln -sf /opt/llama.cpp/build/bin/llama-cli /usr/local/bin/llama-cli && cat

/usr/local/bin/chat-deepseek <<'SH'\n#!/usr/bin/env bash\nmkdir -p

~/.cache/local_llm/history\nrlwrap -a -H ~/.cache/local_llm/history/deepseek.hist llama-cli -m /opt/gguf_models/deepseek-coder-6.7b-base.Q4_K_M.gguf -t \"$(nproc)\" -c 4096 --instruct --interactive-first\nSH\nchmod +x /usr/local/bin/chat-deepseek && echo "installed: place or copy your deepseek .gguf to /opt/gguf_models/deepseek-coder-6.7b-base.Q4_K_M.gguf and run /usr/local/bin/chat-deepseek"' ——————— #OpenHermes / Mistral (one-liner for Kali/Debian): sudo bash -lc 'apt update && apt install -y git build-essential cmake python3 python3-pip rlwrap jq curl && [ -d /opt/llama.cpp ] || git clone https://github.com/ggerganov/llama.cpp.git /opt/llama.cpp && cd /opt/llama.cpp && cmake -S . -B build -DCMAKE_BUILD_TYPE=Release && cmake --build build -j"$(nproc)" && mkdir -p /opt/gguf_models /root/.cache/local_llm/history && touch /opt/gguf_models/openhermes-2.5-mistral-7b.Q4_K_M.gguf && ln -sf /opt/llama.cpp/build/bin/llama-cli /usr/local/bin/llama-cli && cat

/usr/local/bin/chat-openhermes <<'SH'\n#!/usr/bin/env bash\nmkdir -p

~/.cache/local_llm/history\nrlwrap -a -H ~/.cache/local_llm/history/openhermes.hist llama-cli -m /opt/gguf_models/openhermes-2.5-mistral-7b.Q4_K_M.gguf -t \"$(nproc)\" -c 4096 --instruct --interactive-first\nSH\nchmod +x /usr/local/bin/chat-openhermes && echo "installed: place or copy your openhermes .gguf to /opt/gguf_models/openhermes-2.5-mistral-7b.Q4_K_M.gguf and run /usr/local/bin/chat-openhermes"' ————————- #WizardLM (one-liner for Kali/Debian): sudo bash -lc 'apt update && apt install -y git build-essential cmake python3 python3-pip rlwrap jq curl && [ -d /opt/llama.cpp ] || git clone https://github.com/ggerganov/llama.cpp.git /opt/llama.cpp && cd /opt/llama.cpp && cmake -S . -B build -DCMAKE_BUILD_TYPE=Release && cmake --build build -j"$(nproc)" && mkdir -p /opt/gguf_models /root/.cache/local_llm/history && touch /opt/gguf_models/WizardLM-7B-uncensored.Q4_K_M.gguf && ln -sf /opt/llama.cpp/build/bin/llama-cli /usr/local/bin/llama-cli && cat

/usr/local/bin/chat-wizardlm <<'SH'\n#!/usr/bin/env bash\nmkdir -p

~/.cache/local_llm/history\nrlwrap -a -H ~/.cache/local_llm/history/wizardlm.hist llama-cli -m /opt/gguf_models/WizardLM-7B-uncensored.Q4_K_M.gguf -t \"$(nproc)\" -c 4096 --instruct --interactive-first\nSH\nchmod +x /usr/local/bin/chat-wizardlm && echo "installed: place or copy your WizardLM .gguf to /opt/gguf_models/WizardLM-7B-uncensored.Q4_K_M.gguf and run /usr/local/bin/chat-wizardlm"' ————- ### I used the term “one liner” loosely 🤷🏻‍♂️ # I have my training scripts and settings for each model and bash if you decide to install one of these or all of them and when you use my settings to have something to compare notes with. Deepseek really needs to be dialed in to write useable code. I tell Wizard and Hermes what I want Deepseek to do. They then provide a copy paste prompt worded the way Deepseek prefers. Deepseek is NOT conversational. (Wild but true, The Chinese included dark web data breach data in Deepseek coders training. Sometimes it just knows things that it shouldn’t. I asked for a THCHydra prompt to crack a Gmail password. The “assistant” replied: Cum2013 It worked. I blew my mind so I looked into it and that’s how I found out that it was trained on breach data. I hope this helps. Cheers

…

On Tue, Sep 30, 2025 at 1:13 PM endolith ***@***.***> wrote: I just use Open Interpreter <https://github.com/OpenInterpreter/open-interpreter> to effectively replace the shell altogether (but that's also somewhat dead; trying to revive it because I don't know of anything else like it) — Reply to this email directly, view it on GitHub <#725 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BLNA4WKRL3CF3OCDPDT2W4T3VK25PAVCNFSM6AAAAACH35BTB6VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTINJVGUZTAMI> . You are receiving this because you commented.Message ID: ***@***.***>

[Sharon-Needles]

FYI When the Nanny State Media Complex sheeple bleat morality to us from the glass tower of superiorityabout “ fraud GPT” and “ghost GPT“. These unfiltered models are all that they are. They just have the names changed. —/-/———- On Tue, Sep 30, 2025 at 2:43 PM Jessica Paige Youngblood < ***@***.***> wrote:

…

Idk which OS you are using, so I will assume it’s a Debian. So I wrote three install one liners for each of the local model that I’m running. All uncensored and unfiltered. (I’m running Blackarch so if you want that one I will spin it up for you to save you some time.) ———— #DeepSeek Coder (one-liner for Kali/Debian): sudo bash -lc 'apt update && apt install -y git build-essential cmake python3 python3-pip rlwrap jq curl && [ -d /opt/llama.cpp ] || git clone https://github.com/ggerganov/llama.cpp.git /opt/llama.cpp && cd /opt/llama.cpp && cmake -S . -B build -DCMAKE_BUILD_TYPE=Release && cmake --build build -j"$(nproc)" && mkdir -p /opt/gguf_models /root/.cache/local_llm/history && touch /opt/gguf_models/deepseek-coder-6.7b-base.Q4_K_M.gguf && ln -sf /opt/llama.cpp/build/bin/llama-cli /usr/local/bin/llama-cli && cat >/usr/local/bin/chat-deepseek <<'SH'\n#!/usr/bin/env bash\nmkdir -p ~/.cache/local_llm/history\nrlwrap -a -H ~/.cache/local_llm/history/deepseek.hist llama-cli -m /opt/gguf_models/deepseek-coder-6.7b-base.Q4_K_M.gguf -t \"$(nproc)\" -c 4096 --instruct --interactive-first\nSH\nchmod +x /usr/local/bin/chat-deepseek && echo "installed: place or copy your deepseek .gguf to /opt/gguf_models/deepseek-coder-6.7b-base.Q4_K_M.gguf and run /usr/local/bin/chat-deepseek"' ——————— #OpenHermes / Mistral (one-liner for Kali/Debian): sudo bash -lc 'apt update && apt install -y git build-essential cmake python3 python3-pip rlwrap jq curl && [ -d /opt/llama.cpp ] || git clone https://github.com/ggerganov/llama.cpp.git /opt/llama.cpp && cd /opt/llama.cpp && cmake -S . -B build -DCMAKE_BUILD_TYPE=Release && cmake --build build -j"$(nproc)" && mkdir -p /opt/gguf_models /root/.cache/local_llm/history && touch /opt/gguf_models/openhermes-2.5-mistral-7b.Q4_K_M.gguf && ln -sf /opt/llama.cpp/build/bin/llama-cli /usr/local/bin/llama-cli && cat >/usr/local/bin/chat-openhermes <<'SH'\n#!/usr/bin/env bash\nmkdir -p ~/.cache/local_llm/history\nrlwrap -a -H ~/.cache/local_llm/history/openhermes.hist llama-cli -m /opt/gguf_models/openhermes-2.5-mistral-7b.Q4_K_M.gguf -t \"$(nproc)\" -c 4096 --instruct --interactive-first\nSH\nchmod +x /usr/local/bin/chat-openhermes && echo "installed: place or copy your openhermes .gguf to /opt/gguf_models/openhermes-2.5-mistral-7b.Q4_K_M.gguf and run /usr/local/bin/chat-openhermes"' ————————- #WizardLM (one-liner for Kali/Debian): sudo bash -lc 'apt update && apt install -y git build-essential cmake python3 python3-pip rlwrap jq curl && [ -d /opt/llama.cpp ] || git clone https://github.com/ggerganov/llama.cpp.git /opt/llama.cpp && cd /opt/llama.cpp && cmake -S . -B build -DCMAKE_BUILD_TYPE=Release && cmake --build build -j"$(nproc)" && mkdir -p /opt/gguf_models /root/.cache/local_llm/history && touch /opt/gguf_models/WizardLM-7B-uncensored.Q4_K_M.gguf && ln -sf /opt/llama.cpp/build/bin/llama-cli /usr/local/bin/llama-cli && cat >/usr/local/bin/chat-wizardlm <<'SH'\n#!/usr/bin/env bash\nmkdir -p ~/.cache/local_llm/history\nrlwrap -a -H ~/.cache/local_llm/history/wizardlm.hist llama-cli -m /opt/gguf_models/WizardLM-7B-uncensored.Q4_K_M.gguf -t \"$(nproc)\" -c 4096 --instruct --interactive-first\nSH\nchmod +x /usr/local/bin/chat-wizardlm && echo "installed: place or copy your WizardLM .gguf to /opt/gguf_models/WizardLM-7B-uncensored.Q4_K_M.gguf and run /usr/local/bin/chat-wizardlm"' ————- ### I used the term “one liner” loosely 🤷🏻‍♂️ # I have my training scripts and settings for each model and bash if you decide to install one of these or all of them and when you use my settings to have something to compare notes with. Deepseek really needs to be dialed in to write useable code. I tell Wizard and Hermes what I want Deepseek to do. They then provide a copy paste prompt worded the way Deepseek prefers. Deepseek is NOT conversational. (Wild but true, The Chinese included dark web data breach data in Deepseek coders training. Sometimes it just knows things that it shouldn’t. I asked for a THCHydra prompt to crack a Gmail password. The “assistant” replied: Cum2013 It worked. I blew my mind so I looked into it and that’s how I found out that it was trained on breach data. I hope this helps. Cheers On Tue, Sep 30, 2025 at 1:13 PM endolith ***@***.***> wrote: > I just use Open Interpreter > <https://github.com/OpenInterpreter/open-interpreter> to effectively > replace the shell altogether (but that's also somewhat dead; trying to > revive it because I don't know of anything else like it) > > — > Reply to this email directly, view it on GitHub > <#725 (comment)>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/BLNA4WKRL3CF3OCDPDT2W4T3VK25PAVCNFSM6AAAAACH35BTB6VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTINJVGUZTAMI> > . > You are receiving this because you commented.Message ID: > ***@***.***> >

[DanijelDomazet]

It works, no?
Is there something you need that currently isn't supported?

[ColtonKinstley]

I mean yeah it does actually. Still running well on gpt-4o and useable for sure. I was hoping to take advantage of the 5 class models. The thing is, you don't only need fix the --temperature issue. In order to run these efficiently, we really need to turn off thinking, which isn't currently an option.

[DanijelDomazet]

Didn't know about the "thinking" mode. But I did some changes on my own, some time ago I forked this project, and enabled the O3 usage, also added a few more things I was missing. Have a look: https://github.com/DanijelDomazet/hell_gpt.

Let me know about the "thinking" feature, might be worth looking into it.