Skip to content

Environment Variables

Environment variables override both global and workspace config.toml settings for the current terminal process session.


VariableConfig KeyMeaning / Usage
CTXSIFT_MAX_OUTPUT_TOKENSmax_output_tokensLimit output token length.
CTXSIFT_TIMEOUT_MStimeout_msRequest timeout duration in milliseconds.
CTXSIFT_RETRIESretriesRequest retry count.
CTXSIFT_RECOVERY_ENABLEDrecovery_enabledEnable or disable deterministic output recovery before final return.
VariableConfig KeyMeaning / Usage
CTXSIFT_LOCAL_MODELlocal.modelHugging Face model repository ID.
CTXSIFT_LOCAL_GGUF_FILENAMElocal.gguf_filenameGGUF filename (required for CPU llama.cpp).
CTXSIFT_LOCAL_LLAMA_CONTEXT_WINDOWlocal.llama_context_windowCPU context window size.
CTXSIFT_LOCAL_DEVICElocal.deviceExecution device: auto, cpu, cuda, mps.
CTXSIFT_LOCAL_DTYPElocal.dtypeGPU model precision: auto, float16, bfloat16.
CTXSIFT_LOCAL_ATTN_IMPLEMENTATIONlocal.attn_implementationAttention implementation: sdpa, flash_attention_2.
CTXSIFT_LOCAL_QUANTIZATIONlocal.quantizationQuantization: none, bnb-8bit, bnb-4bit-fp4, bnb-4bit-nf4.
CTXSIFT_MODEL_CACHE_PATHlocal.model_cache_pathCustom quantized model storage path.
VariableConfig KeyMeaning / Usage
CTXSIFT_LLM_BASE_URLremote.base_urlLiteLLM-compatible provider API base URL.
CTXSIFT_LLM_MODELremote.model_nameRemote model name identifier (e.g. gpt-4o-mini).
CTXSIFT_LLM_API_KEYremote.api_keyAPI Key for remote authentication.
CTXSIFT_LLM_API_VERSIONremote.api_versionOptional provider API version string.
CTXSIFT_LLM_REASONING_MODEremote.reasoning_modeReasoning settings: auto, true, false.
VariableConfig KeyMeaning / Usage
CTXSIFT_EMBEDDING_MODELembedding.modelSentence Transformers model ID.
CTXSIFT_EMBEDDING_BACKENDembedding.backendExecution engine: auto, onnx, torch.
CTXSIFT_EMBEDDING_DEVICEembedding.deviceEmbedding hardware device: auto, cpu, cuda.
CTXSIFT_EMBEDDING_DTYPEembedding.dtypeEmbedding model precision.
CTXSIFT_EMBEDDING_ATTN_IMPLEMENTATIONembedding.attn_implementationAttention mechanism backend.
CTXSIFT_EMBEDDING_MAX_LENGTHembedding.max_lengthMaximum sequence context limit.
CTXSIFT_EMBEDDING_QUERY_PROMPT_NAMEembedding.query_prompt_namePreset query prompt template.
CTXSIFT_EMBEDDING_QUERY_PROMPTembedding.query_promptCustom query prompt prefix string.
CTXSIFT_EMBEDDING_DOCUMENT_PROMPT_NAMEembedding.document_prompt_namePreset document prompt template.
VariableConfig KeyMeaning / Usage
CTXSIFT_RECALL_DEFAULT_LIMITrecall.default_limitDefault number of recall results returned.
CTXSIFT_RECALL_LEXICAL_CANDIDATE_LIMITrecall.lexical_candidate_limitMax lexical candidate matches queried.
CTXSIFT_RECALL_VECTOR_CANDIDATE_LIMITrecall.vector_candidate_limitMax semantic vector candidate matches queried.
CTXSIFT_RECALL_MAX_VECTOR_DISTANCErecall.max_vector_distanceCosine distance similarity filter limit.
VariableConfig KeyMeaning / Usage
CTXSIFT_DAEMON_ENABLEDdaemon.enabledEnable background serving: true, false.
CTXSIFT_DAEMON_IDLE_TIMEOUT_SECONDSdaemon.idle_timeout_secondsShutdown inactivity window in seconds.
CTXSIFT_DAEMON_STARTUP_TIMEOUT_MSdaemon.startup_timeout_msMaximum boot wait time in milliseconds.
CTXSIFT_DAEMON_EMBEDDING_BATCH_WINDOW_MSdaemon.embedding_batch_window_msRequests grouping latency window.
CTXSIFT_DAEMON_EMBEDDING_MAX_BATCH_SIZEdaemon.embedding_max_batch_sizeLimit of grouped embedding requests.
VariableConfig KeyMeaning / Usage
CTXSIFT_RETENTION_MAX_AGE_DAYSretention.max_age_daysNumber of days to keep compressed records.