Environment Variables

Set these in the shell, a .env file, or the process environment. A .env.example is included.

Connection

Variable	Required	Default	Description
`OPENAI_API_KEY`	no	`""`	API key
`OPENAI_BASE_URL`	no	`https://api.openai.com/v1`	API endpoint (this is the OpenAI SDK default; when using a local server like LM Studio, set this to `http://localhost:1234/v1`)
`LLM_CHAT_OPENAI_DEFAULT_MODEL`	no	`""`	Fallback model name

Read at OpenAIChatService construction:

Variable	Description
`LLM_CHAT_OPENAI_TEMPERATURE`	Sampling temperature (0-2)
`LLM_CHAT_OPENAI_MAX_TOKENS`	`max_tokens`
`LLM_CHAT_OPENAI_MAX_COMPLETION_TOKENS`	`max_completion_tokens` (o1/o3, takes precedence)
`LLM_CHAT_OPENAI_TOP_P`	Nucleus sampling
`LLM_CHAT_OPENAI_REASONING_EFFORT`	`reasoning_effort` for o-series models (`none`, `minimal`, `low`, `medium`, `high`, `xhigh`)
`LLM_CHAT_OPENAI_TOOL_CHOICE`	`tool_choice` (`none`, `auto`, `required`)
`LLM_CHAT_OPENAI_VERBOSITY`	`verbosity` (`low`, `medium`, `high`)
`LLM_CHAT_OPENAI_INCLUDE_USAGE`	Request streamed token usage via `stream_options.include_usage` (`true`/`1`; default `false`)

Variable	Default	Description
`LLM_CHAT_MAX_TOOL_CALL_ROUNDS`	`10`	Max tool-call recursion depth
`LLM_CHAT_CONTEXT_WINDOW_SIZE_HINT`	`0`	Context window size hint in tokens (`0` = unlimited). Used by `isContextWindowExceeded()`

Variable	Default	Description
`LLM_CHAT_SYSTEM_PROMPT_DIR`	`./prompts/`	Directory containing `.md` and `.txt` system prompt files; each file becomes a component under `general`. Read only if it exists; never created
`LLM_CHAT_USER_PROMPTS`	`""`	Comma-separated file paths for initial user messages

Variable	Default	Description
`LLM_CHAT_HOOKS_DIR`	none	Directory containing `.json` hook definition files (auto-loaded on init/send)