Configuration

OpenSyntax reads saved config, provider credentials, environment variables, and CLI flags. CLI flags win for the current run.

Config locations

File	Purpose
`~/.opensyntax/config.json`	Runtime settings such as provider, model, permission, temperature, and token limits.
`~/.opensyntax/providers.json`	Connected providers and encrypted local credentials.
`~/.opensyntax/sessions/`	Saved conversations, plans, and tool logs.

Provider setup wizard

opensyntax auth

The wizard asks for provider, authentication method, API key when needed, base URL for local or Azure providers, default model, and optional validation.

Direct config command

opensyntax config \
  --provider openai \
  --model gpt-4o-mini \
  --api-key "$OPENAI_API_KEY" \
  --permission shell-safe

Environment variables

export OPENSYNTAX_PROVIDER=nvidia
export OPENSYNTAX_MODEL=meta/llama-3.1-70b-instruct
export OPENSYNTAX_API_KEY=nvapi-...
export OPENSYNTAX_BASE_URL=https://integrate.api.nvidia.com/v1
export OPENSYNTAX_TEMPERATURE=0.2
export OPENSYNTAX_MAX_TOKENS=4096

Permission levels

Level	Behavior
`read-only`	Read/search/git inspection only.
`workspace-safe`	Allows tests, builds, lint, typecheck, and read-only git/package-manager commands.
`workspace-write`	Allows workspace edits and dependency installs with approval.
`shell-safe`	Allows developer shell commands and prompts for installs, network, and risky commands.
`full-os`	Requires approval for OS-level commands such as package manager system installs and service changes.
`danger`	Requires typed confirmation for destructive commands.

Terminal config

{
  "commandTimeoutMs": 120000,
  "allowFullOsCommands": false,
  "requireApprovalForInstall": true,
  "requireApprovalForNetwork": true,
  "shellMode": "shell-safe"
}