Configuration reference
The server (stromboli serve) is configured only with environment variables. .env.example lists them with comments. Durations use Go syntax (30s, 5m, 1h30m); booleans accept true/false, 1/0, t/f.
The server refuses to start when a value cannot be parsed or a required setting is missing. Settings that are half-configured (Slack buttons without actions, a summary without health checks…) only log a warning.
Server
| Variable |
Default |
|
STROMBOLI_ADDR |
:8080 |
listen address. Plain HTTP/1.1 and unencrypted HTTP/2 (for gRPC clients) |
STROMBOLI_DB_URL |
required |
Postgres URL. Migrations run on startup |
STROMBOLI_SECRET_KEY |
required |
encrypts connector secrets with AES-256-GCM (the key is the SHA-256 of this value). Must never change once connectors exist. Generate with openssl rand -base64 32 |
Authentication
| Variable |
Default |
|
STROMBOLI_API_TOKEN |
empty = no auth |
bearer token for the API. Required by STROMBOLI_ACTIONS_ENABLED and STROMBOLI_MCP_ENABLED |
STROMBOLI_WEBHOOK_TOKEN |
STROMBOLI_API_TOKEN |
token for /webhooks/*, sent as ?token=, Authorization: Bearer or X-Webhook-Secret. Empty (with no API token) = no auth |
Models and providers
| Variable |
Default |
|
STROMBOLI_MODEL |
anthropic:claude-opus-5-5 |
default model for every task, as provider:model |
STROMBOLI_EFFORT |
provider default |
low, medium, high, xhigh or max |
STROMBOLI_TRIAGE_MODEL |
STROMBOLI_MODEL |
model until the mitigation is published |
STROMBOLI_TRIAGE_EFFORT |
STROMBOLI_EFFORT |
|
STROMBOLI_INVESTIGATE_MODEL |
STROMBOLI_MODEL |
model for the root cause and the report; takes over the triage conversation |
STROMBOLI_INVESTIGATE_EFFORT |
STROMBOLI_EFFORT |
|
Model ids:
| Id |
Provider |
Variables |
anthropic:<model>, or a bare claude-… id |
Anthropic Messages API |
ANTHROPIC_API_KEY |
openai:<model> |
OpenAI Chat Completions |
OPENAI_API_KEY, OPENAI_BASE_URL (default https://api.openai.com/v1) |
google:<model> |
Gemini, OpenAI-compatible endpoint |
GEMINI_API_KEY |
openrouter:<model> |
OpenRouter |
OPENROUTER_API_KEY |
deepseek:<model> |
DeepSeek |
DEEPSEEK_API_KEY, DEEPSEEK_BASE_URL (default https://api.deepseek.com) |
gateway:<id>, or any <vendor>/<model> id |
Vercel AI Gateway |
AI_GATEWAY_API_KEY or VERCEL_OIDC_TOKEN, AI_GATEWAY_BASE_URL |
ollama:<model> |
Ollama |
OLLAMA_BASE_URL (default http://localhost:11434/v1) |
<name>:<model> |
any OpenAI-compatible API |
STROMBOLI_PROVIDER_<NAME>_BASE_URL (required), STROMBOLI_PROVIDER_<NAME>_API_KEY. <NAME> is the name upper-cased, - as _ |
If a model cannot be resolved, the server starts with investigations disabled; briefs still work.
Provider behaviour: Anthropic models (direct or through the gateway’s Anthropic Messages API) use prompt caching and effort. Other providers use the OpenAI Chat Completions format, where effort is sent as reasoning_effort, only when set. Reasoning a provider returns (DeepSeek’s reasoning_content, the gateway’s reasoning_details) is sent back to it on later turns. Switching models between tasks resets the prompt cache once.
Investigations
| Variable |
Default |
|
STROMBOLI_MAX_TOOL_CALLS |
30 |
tool calls per investigation. When reached, the agent is told to submit; it fails after 10 more |
STROMBOLI_MITIGATION_TOOL_CALLS |
6 |
tool calls after which the agent is told to submit a mitigation now |
STROMBOLI_INVESTIGATION_TIMEOUT |
8m |
wall-clock limit of one investigation |
STROMBOLI_DEDUPE_WINDOW |
30m |
how long after it finished an investigation still absorbs repeats of its alert (webhooks) |
Machine actions
| Variable |
Default |
|
STROMBOLI_ACTIONS_ENABLED |
false |
let investigations propose machine actions a human can approve. Requires STROMBOLI_API_TOKEN |
STROMBOLI_ACTION_TTL |
30m |
how long a proposed action can be approved. Must be positive |
Health checks
| Variable |
Default |
|
STROMBOLI_HEALTHCHECK_INTERVAL |
empty = off |
how often to check Fly metrics against thresholds, at least 5m. 0 disables |
STROMBOLI_HEALTHCHECK_SUMMARY |
false |
post an all-clear to Slack after each run where every service is healthy, and a warning listing the services that could not be checked. Needs Slack and health checks |
Slack
| Variable |
Default |
|
SLACK_BOT_TOKEN |
|
bot token (xoxb-…) with chat:write. Slack is off unless this and SLACK_CHANNEL are set |
SLACK_CHANNEL |
|
channel name (#incidents) or ID (C0123…) |
SLACK_SIGNING_SECRET |
|
enables Apply/Reject buttons at POST /slack/interactions. Needs Slack and STROMBOLI_ACTIONS_ENABLED=true |
SLACK_APPROVERS |
empty = anyone in the channel |
comma-separated Slack user IDs allowed to click Apply/Reject |
MCP server
| Variable |
Default |
|
STROMBOLI_MCP_ENABLED |
false |
serve the MCP tools over HTTP at /mcp. Requires STROMBOLI_API_TOKEN |
STROMBOLI_MCP_TOKEN |
STROMBOLI_API_TOKEN |
when set, the only token /mcp accepts; it does not reach the rest of the API |
Connector credentials from the environment
Connectors are configured through the API and stored in Postgres, not in the environment. Two exceptions, read by the server process:
cloudwatch without keys uses the default AWS credential chain (AWS_ACCESS_KEY_ID, AWS_PROFILE, an IAM role…).
gcp_logging and gcp_monitoring without credentials_json use Application Default Credentials (GOOGLE_APPLICATION_CREDENTIALS, workload identity…).
Docker Compose only
Read by docker-compose.yml, not by stromboli:
| Variable |
Default |
|
STROMBOLI_VERSION |
latest |
image tag of ghcr.io/openstatushq/stromboli |
STROMBOLI_PORT |
8080 |
host port |
POSTGRES_PASSWORD |
stromboli |
Postgres password; only applied when the volume is created |
CLI
The client commands read STROMBOLI_URL, STROMBOLI_API_TOKEN, STROMBOLI_USER and STROMBOLI_DEBUG: see the CLI reference.
Edit this page on GitHub ↗