reference
Environment Variables
All environment variables supported by the Ninetrix CLI and agent runtime.
#LLM providers
| Variable | Provider | Description |
|---|
ANTHROPIC_API_KEY | Anthropic | API key for Claude models |
OPENAI_API_KEY | OpenAI | API key for GPT models |
GEMINI_API_KEY | Google | API key for Gemini models |
DEEPSEEK_API_KEY | DeepSeek | API key for DeepSeek models |
MISTRAL_API_KEY | Mistral | API key for Mistral models |
GROQ_API_KEY | Groq | API key for Groq-hosted models |
TOGETHERAI_API_KEY | Together AI | API key for Together AI inference |
OPENROUTER_API_KEY | OpenRouter | API key for OpenRouter model routing |
CEREBRAS_API_KEY | Cerebras | API key for Cerebras inference |
FIREWORKS_API_KEY | Fireworks | API key for Fireworks AI inference |
AWS_ACCESS_KEY_ID | AWS Bedrock | AWS access key for Bedrock models |
AWS_SECRET_ACCESS_KEY | AWS Bedrock | AWS secret key for Bedrock models |
AWS_REGION_NAME | AWS Bedrock | AWS region for Bedrock (e.g. us-east-1) |
AZURE_API_KEY | Azure OpenAI | API key for Azure OpenAI deployment |
AZURE_API_BASE | Azure OpenAI | Azure OpenAI resource endpoint URL |
AZURE_API_VERSION | Azure OpenAI | Azure OpenAI API version (e.g. 2024-02-01) |
MINIMAX_API_KEY | MiniMax | API key for MiniMax models |
#Runtime overrides
These override the baked-in values from agentfile.yaml at runtime — no rebuild required:
| Variable | Overrides | Example |
|---|
AGENTFILE_PROVIDER | runtime.provider | openai |
AGENTFILE_MODEL | runtime.model | gpt-4o |
AGENTFILE_TEMPERATURE | runtime.temperature | 0.5 |
AGENTFILE_THINKING_ENABLED | execution.thinking.enabled | true |
AGENTFILE_MAX_TOKENS | runtime.max_tokens | 8192 |
AGENTFILE_MAX_TURNS | runtime.max_turns | 20 |
AGENTFILE_TOOL_TIMEOUT | runtime.tool_timeout | 30 |
AGENTFILE_HISTORY_WINDOW_TOKENS | runtime.history_window_tokens | 90000 |
AGENTFILE_SYSTEM_PROMPT | Entire system prompt | You are a helpful assistant. |
AGENTFILE_VERIFIER_MODEL | execution.verifier.model | claude-haiku-4-5-20251001 |
#Agent runtime
These control agent behavior inside the running container:
| Variable | Description | Default |
|---|
AGENTFILE_THREAD_ID | Set the thread ID for the run. Used for session persistence and resume. | Auto-generated UUID |
AGENTFILE_SHOW_COST | Print per-call USD cost to stdout. Set to 1, true, or yes to enable. | false |
AGENTFILE_APPROVAL_ENABLED | Toggle HITL approval gate on/off at runtime. Set to false, 0, or no to disable. | true |
#CLI
| Variable | Description |
|---|
NINETRIX_DEBUG | Set to 1 to show full Python tracebacks on CLI errors instead of friendly messages. |
#Telemetry
| Variable | Description |
|---|
AGENTFILE_API_URL | URL of the API server (local or cloud). Set by ninetrix dev automatically. |
AGENTFILE_RUNNER_TOKEN | Bearer token for the API server. Set from ~/.agentfile/.api-secret. |
#Multi-agent
| Variable | Description |
|---|
AGENTFILE_PEER_{NAME}_URL | URL of a peer agent. Injected by ninetrix up. e.g. AGENTFILE_PEER_RESEARCHER_URL=http://researcher:9000 |
#MCP tools
| Variable | Description |
|---|
MCP_GATEWAY_URL | URL of the MCP gateway (default: http://localhost:9090) |
MCP_GATEWAY_SECRET | Shared secret for MCP gateway auth (default: dev-secret in local mode) |
GITHUB_TOKEN | GitHub personal access token for the GitHub MCP server |
BRAVE_API_KEY | Brave Search API key for the Brave Search MCP server |
SLACK_BOT_TOKEN | Slack bot token for the Slack MCP server |
NOTION_API_KEY | Notion API key for the Notion MCP server |
