CALIBER
Quickstart
Reference

Config and environment reference

The operator and integrator lookup surface for the configuration families exposed in the current repository.

OperatorDeveloperReferenceGA
PrerequisitesA CALIBER deployment or integration configuration question
Reviewed 2026-08-10 · current main branch docs contract

This page is the operator and integrator lookup surface for the configuration families that are exposed in the current repository. Use it as the fast index, then move into the deeper runtime docs when you need subsystem detail.

At a glance

FamilyRepresentative settingsUse when
Local ports and launchCALIBER_PORT, MLFLOW_PORT, MLFLOW_GATEWAY_PORTbringing up the local suite
Core runtimeCALIBER_DATABASE_URL, CALIBER_LOG_SINKwiring the control plane and logging
LLM provider and gatewayCALIBER_LLM_PROVIDER, CALIBER_LLM_BASE_URLchoosing provider routing
Object storeCALIBER_OBJECT_STORE_*uploads, extraction, previews, log mirroring
Workflow storageCALIBER_WORKFLOW_STORAGE_*run workspaces, limits, retention, signed URLs
Event backendCALIBER_WORKFLOW_RUN_EVENT_BACKENDworkflow queue and runtime eventing
Deep reference · data models, APIs & lifecycle

Reference

1. Local launch and ports

SettingPurpose
CALIBER_PORThost port for the CALIBER UI/API service
MLFLOW_PORThost port for the MLflow UI; start.sh guards common collisions
MLFLOW_GATEWAY_PORThost port for the MLflow AI Gateway surface

2. Core runtime

SettingPurpose
CALIBER_DATABASE_URLSQLAlchemy URL for the CALIBER metadata store
CALIBER_LOG_SINKchoose stderr-only or S3/object-store-backed JSONL mirroring
CALIBER_STATIC_PREFIXserve behind a reverse-proxy prefix when applicable

3. LLM and gateway

SettingPurpose
CALIBER_LLM_PROVIDERselect the default provider implementation
CALIBER_LLM_DIAGNOSIS_MODELdefault OpenAI model; gpt-5.6-luna when unset
CALIBER_LLM_REASONING_EFFORTplatform-wide reasoning default; high when unset
CALIBER_GEPA_REFLECTION_MODELGEPA reflection model; inherits the Luna default
CALIBER_MEMORY_LLM_MODELmemory extraction model; inherits the Luna default
CALIBER_LLM_BASE_URLoverride the OpenAI-compatible base URL
CALIBER_GATEWAY_URIdiscover an MLflow AI Gateway surface

The assistant uses the same OpenAI default through CALIBER_ASSISTANT_MODEL and CALIBER_ASSISTANT_REASONING. Explicit per-asset and per-call selections remain supported overrides.

4. Object store

SettingPurpose
CALIBER_OBJECT_STORE_ENDPOINT_URLobject-store endpoint
CALIBER_OBJECT_STORE_REGIONobject-store region
CALIBER_OBJECT_STORE_ACCESS_KEY_SOURCEsecret source for access key
CALIBER_OBJECT_STORE_SECRET_KEY_SOURCEsecret source for secret key
CALIBER_OBJECT_STORE_FORCE_PATH_STYLEpath-style S3/MinIO behavior

5. Workflow storage

SettingPurpose
CALIBER_WORKFLOW_STORAGE_BACKENDlocal or s3
CALIBER_WORKFLOW_STORAGE_BASE_URIlocal workspace root when using local backend
CALIBER_WORKFLOW_STORAGE_BUCKETS3/MinIO bucket for workflow files
CALIBER_WORKFLOW_STORAGE_PREFIXkey prefix for workflow files
CALIBER_WORKFLOW_STORAGE_INTERNAL_ENDPOINT_URLinternal endpoint for worker-side access
CALIBER_WORKFLOW_STORAGE_PUBLIC_ENDPOINT_URLpublic endpoint for signed URLs
CALIBER_WORKFLOW_STORAGE_MAX_UPLOAD_BYTESper-upload limit
CALIBER_WORKFLOW_STORAGE_MAX_RUN_BYTESper-run total byte limit
CALIBER_WORKFLOW_STORAGE_MAX_FILES_PER_RUNper-run file-count limit
CALIBER_WORKFLOW_STORAGE_SIGNED_URL_TTL_SECONDSdefault signed URL lifetime
CALIBER_WORKFLOW_STORAGE_RETENTION_*run-artifact retention policy

6. Event backend

SettingPurpose
CALIBER_WORKFLOW_RUN_EVENT_BACKENDworkflow event transport/backing mode

7. Secret-source convention

Prefer settings ending in *_SOURCE for credentials and sensitive values. The current runtime is designed around secret sources rather than spreading literal keys throughout deployment files.

CALIBER : Contextual Adaptive Lifecycle for Intelligent Build, Evaluation, and Refinement — this page is generated from the authoritative Markdown sources in docs/ and the repository-level ARCHITECTURE.md.