Documentation

Configuration reference

Runtime configuration is read from environment variables and the repository-local .env file.

Core settings

VariableDefaultPurpose
CCIIPSCOPE_DATA_DIR./dataDatabase, uploads, parsed assets, and logs
CCIIPSCOPE_RETENTION_DAYS30Inactive-project retention period
CCIIPSCOPE_MAX_UPLOAD_BYTES52428800Maximum upload size
CCIIPSCOPE_CLEANUP_INTERVAL21600Cleanup interval in seconds
CCIIPSCOPE_ENABLE_LOCAL_COMPILEfalseAllow host-side LaTeX compilation

Model configuration

The built-in MiniMax profile uses MINIMAX_API_KEY and CCIIPSCOPE_MINIMAX_BASE_URL. Additional configurations use a normalized nickname:

CCIIPSCOPE_MODEL_REVIEW_DISPLAY_NAME=Review Model
CCIIPSCOPE_MODEL_REVIEW_PROVIDER=openai
CCIIPSCOPE_MODEL_REVIEW_MODEL=model-name
CCIIPSCOPE_MODEL_REVIEW_BASE_URL=https://api.example.com/v1
CCIIPSCOPE_MODEL_REVIEW_API_KEY=secret

Provider values identify the client protocol, not the vendor brand. Keep context-window and structured-output behavior in mind when selecting a model: the workflow passes large paper contexts and validates multiple schemas.

MinerU PDF parsing

VariableDefault
CCIIPSCOPE_PDF_PARSERmineru
CCIIPSCOPE_MINERU_API_URLhttp://localhost:8008
CCIIPSCOPE_MINERU_GPU_MODEauto
CCIIPSCOPE_MINERU_GPU_MAX_MEMORY_MB2048
CCIIPSCOPE_MINERU_GPU_MAX_UTILIZATION20
CCIIPSCOPE_MINERU_TIMEOUT1800
CCIIPSCOPE_MINERU_PARSE_BACKENDpipeline

GPU selection occurs when parsing begins. The container is stopped in a finally path after success or failure so GPU memory is released between projects.

VariableDefaultPurpose
CCIIPSCOPE_SEARXNG_URLhttp://localhost:8888General search endpoint
CCIIPSCOPE_ACADEMIC_CONTACTemptyContact in academic-provider user agent
CCIIPSCOPE_SEARCH_CACHE_TTL7 daysProvider-result cache duration

Academic search combines general discovery with arXiv, OpenReview, and DBLP when available. Provider failures are isolated so one rate-limited source does not invalidate the complete landscape.

Apply configuration changes

Restart CciipScope after editing .env. The available model list is shown when the application starts, and the selected model can be changed from a completed workspace.