All my thanks to chatdev i made almost the least change to the codes, this version allows you to use any online-openai-style-api, besides, your local ollama is supported, too. part of the codes are still a mass, i'd pay some time to reconstruct it(when i'm not lazy though)
1.Create an env for the project, as my local one:python3.9, created by conda, on my windows i just type
conda create --prefix A:\Pvenv\conda_p39_chatdev python=3.9
conda activate A:\Pvenv\conda_p39_chatdev
2.clone my git, you should just do
git clone https://github.com/skypu/ChatDev.git
3.install the requirements, you'd better cd to your chatdev dir, make sure you're in the correct conda env and type
pip install -r requirements.txt
4.make sure you have valid api-provider info, and set the enviroments, on my windows i just type
$env:OPENAI_API_KEY="your_openai_api_key"
$env:BASE_URL="https://ark.cn-beijing.volces.com/api/v3"
$env:DEFAULT_AI_MODEL="deepseek-r1-250120"
$env:AI_PROVIDER="openai"
alternatively, you could just use your local ollama with any model you've installed, all calls are free
$env:OPENAI_API_KEY="not needed, but keep it"
$env:BASE_URL="http://localhost:11434/api/generate"
$env:DEFAULT_AI_MODEL="gemma3:latest"
$env:AI_PROVIDER="ollama"
5.just python and enjoy it
python run.py --org "MyGreateCompany" --name "MyFirstProject" --task "I'd like to blablabla"(any idea you have)
model_backend.py:line 69; because the tiktoken module is designed for openai, deepseek models encoding methods are not included, so i set it to "cl100k_base" to use local ollama, i modified the model_backend, added a func call_ai, a class to handel the ollama response:OllamaChatCompletion in other files i just commented all the gpt stuffs.
if you have any problems, feel free to talk with me.
