Configuration Reference
The ICE engine is configured exclusively via environment variables. These parameters govern memory allocation, scaling, security, and document processing.
1. Core Infrastructure
| Variable | Description | Default |
|---|---|---|
ICE_PORT | Port the engine listens on. | 8000 |
ICE_WORKERS | Number of Uvicorn worker processes. | 4 |
DATABASE_URL | PostgreSQL connection string (requires pgvector). | N/A |
REDIS_URL | Redis connection string for Hot-Cache. | N/A |
ICE_LICENSE_JWT | License key for Pro/Enterprise tiers. | N/A |
ICE_UPLOAD_DIR | Sandboxed directory for multimodal file ingestion. | /tmp/ice/uploads |
2. Resource & Scaling
| Variable | Description | Default |
|---|---|---|
ICE_MEMORY_CAP_GB | Hard RAM limit for the ICE process. | 8 |
ICE_MAX_STITCH_CONCURRENCY | Maximum concurrent context operations. | 16 |
ICE_RETENTION_DAYS | Data retention policy (days). | 30 |
ICE_PRE_PURGE_WEBHOOK_URL | Webhook URL for cold storage archiving before deletion. | N/A |
3. Context & Prompt Engineering
| Variable | Description | Default |
|---|---|---|
ICE_MODEL_CONTEXT_WINDOW | Total virtual context budget (input + output). | 128000 |
ICE_POST_COMPRESSION_LIMIT | Maximum final prompt size sent to the LLM. | 100000 |
ICE_DEFAULT_OUTPUT_TOKENS | Default output token budget if not specified. | 4096 |
ICE_MAX_CONTINUATIONS | Maximum recursive turns for Recursive Completion. | 5 |
ICE_DYNAMIC_OUTPUT_BUDGET | Enables intent-based token budgeting. | true |
4. Multimodal Ingestion
| Variable | Description | Default |
|---|---|---|
ICE_CHUNK_SIZE | Size of text chunks for semantic partitioning. | 1024 |
ICE_CHUNK_OVERLAP | Overlap between consecutive chunks. | 128 |
ICE_INGEST_PARSER | Document parser selection (docling, mineru). | docling |
ICE_INGEST_ENABLE_IMAGES | Process and embed images in documents. | false |
ICE_INGEST_ENABLE_TABLES | Convert tables to Markdown during ingestion. | true |
ICE_INGEST_ENABLE_EQUATIONS | Extract and encode mathematical equations. | true |
ICE_INGEST_ENABLE_COMPLIANCE | Enable PII redaction and compliance scrubbing. | false |
ICE_INGEST_COMPLIANCE_LEVEL | Compliance depth (basic, strict, full). | basic |
5. LLM Routing
| Variable | Description | Default |
|---|---|---|
UPSTREAM_API_URL | Global fallback URL for LLM requests. | N/A |
OPENAI_API_URL | Dedicated routing for OpenAI-compatible models. | https://api.openai.com/v1 |
ANTHROPIC_API_URL | Dedicated routing for Anthropic models. | https://api.anthropic.com/v1 |