Configuration

kura reads ~/.kura/config.toml. Edit it directly, or use kura config:

kura config list
kura config get llm.provider
kura config set general.stale_days 90

config.toml

[general]
default_bucket = "main"
stale_days = 180

[llm]
provider = "auto"     # auto | ollama | lmstudio | none

[llm.models]
embedding = "qwen3-embedding:0.6b"
embedding_dimensions = 1024
reranker = "dengcao/Qwen3-Reranker-0.6B"
generation = "qwen3:4b"

[clip]
path = "clips"

[general]

Key	Meaning
default_bucket	Bucket new documents land in when none is given
stale_days	A document untouched this long is a staleness candidate (kura ls --stale)

[llm]

provider selects the local LLM backend:

• auto — try Ollama first, then LM Studio (the default).
• ollama / lmstudio — force one backend.
• none — disable LLM features entirely; kura runs keyword-only.

[llm.models]

The models kura asks the provider for. Defaults are small enough to run together on a 32 GB Mac:

Key	Used for
embedding	Vectors for semantic and hybrid search
embedding_dimensions	Vector width; must match the embedding model
reranker	Reranking in kura query
generation	kura clip cleanup, tag suggestion, query expansion

Caution

After changing the embedding model or its dimensions, run kura doctor --fix (to detect the change) and then kura embed to regenerate every vector at the new dimensions. Old vectors are not compatible across models.

[clip]

Key	Meaning
path	Document path new clips are filed under (default clips); set "" to clip into the bucket root

Environment variables

Variable	Effect
KURA_HOME	Data directory (default ~/.kura)
KURA_DB	Path to the database file, overriding KURA_HOME/kura.db
NO_COLOR	Disable colored CLI output

KURA_HOME is the clean way to keep separate knowledge bases — point it at a different directory per project or context, and each gets its own SQLite store and config.