Skip to content

Latest commit

 

History

History
74 lines (43 loc) · 2.29 KB

File metadata and controls

74 lines (43 loc) · 2.29 KB

Auto-Compose

The nchat bundled default auto-compose utility compose uses external services for chat completion, and generally requires an API key (set in environment) to work. Environment variables to set:

OpenAI: OPENAI_API_KEY
Gemini: GEMINI_API_KEY

Basic Testing

An example chat history file is provided for testing chat completion standalone. Example:

./src/compose -c doc/example-history.txt

Testing Services / Models

The utility compose may be used with OpenAI-compatible services.

Example usage with default service (OpenAI) and default model (gpt-4o-mini):

./src/compose -c doc/example-history.txt

Example usage with OpenAI:

./src/compose -s openai -c doc/example-history.txt

Example usage with OpenAI and custom model and longer timeout of 60 secs:

./src/compose -s openai -m gpt-5-nano -T 60 -c doc/example-history.txt

Example usage with Google Gemini:

./src/compose -s gemini -c doc/example-history.txt

Example usage with Google Gemini and custom model:

./src/compose -s gemini -m gemini-2.5-flash -c doc/example-history.txt

Example usage with llama.cpp:

./src/compose -s "http://192.168.10.159:8080" -c doc/example-history.txt

Example starting llama.cpp server:

llama-server --port 8080 -m ./models/llama-2-13b-chat.Q4_K_M.gguf

Configuring Custom Service / Model

Edit ui.conf to match the desired compose path and usage.

The compose script is bundled inside the nchat executable and extracted to a temporary location at runtime, so there is no installed copy to reference. To use custom arguments, copy compose from the source tree (or download it) to a location of your choice, e.g:

cp src/compose ~/.local/bin/compose

Example usage with Google Gemini and custom model:

auto_compose_command=python3 ~/.local/bin/compose -s gemini -m gemini-2.0-flash -c '%1'

Example usage with OpenAI and custom model and longer timeout of 60 secs:

auto_compose_command=python3 ~/.local/bin/compose -s openai -m gpt-5-nano -T 60 -c '%1'

Example usage with custom prompt and max token limit of 100:

auto_compose_command=python3 ~/.local/bin/compose -p "Suggest {your_name}'s next reply in a joking manner." -M 100 -c '%1'