hallouminate 0.2.0

A markdown corpus indexer for LLMs to build and query their own per-repo wikis.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51

# Hallouminate configuration. Edit corpus and repository entries to point at
# the markdown trees and repositories you want indexed.

# [[corpus]]
# name = "claude-config"
# paths = ["~/.claude/skills", "~/.claude/agents", "~/.claude/CLAUDE.md"]
# globs = ["**/*.md"]
# exclude = ["**/.git/**", "**/node_modules/**"]

# [[repository]]
# name = "tern"
# path = "~/Dev/tern"
# # Optional source-document corpus; derived as `repo:tern:corpus`.
# # corpus_paths = ["docs"]
# # corpus_globs = ["**/*.md"]
# # corpus_exclude = ["**/drafts/**"]
# # The LLM-managed wiki is always derived as `repo:tern:wiki` and lives
# # at `<path>/.hallouminate/wiki`.
# # repo:tern:code is reserved for a future code-aware indexing slice.

[search]
top_files_default       = 10
chunks_per_file_default = 3
# Optional crossencoder rerank model — runs after FTS+vector+rg fusion
# to re-score the top candidates. Leave commented to skip the step.
# Supported: jina-reranker-v1-turbo-en, jina-reranker-v2-base-multiligual,
# bge-reranker-base, bge-reranker-v2-m3.
# crossencoder = "jina-reranker-v1-turbo-en"

# Dense embeddings are on by default: hallouminate downloads the embedding
# model on first use and fuses the dense (vector) signal with lexical search
# (full-text search + ripgrep + rerank). Set `enabled = false` to retrieve
# lexically only — no embedding model is downloaded, just the tokenizer used
# for chunking.
[embeddings]
enabled   = true
# Supported 384-dim models:
#   snowflake/snowflake-arctic-embed-s (default, retrieval-tuned English)
#   BAAI/bge-small-en-v1.5             (English, retrieval-tuned)
#   intfloat/multilingual-e5-small     (multilingual; no quantized variant)
model     = "snowflake/snowflake-arctic-embed-s"
# Use the quantized (*Q) fastembed variant when one exists. No effect for
# multilingual-e5-small (it ships no quantized ONNX — would error if enabled).
quantized = false
cache_dir = "~/.cache/hallouminate/fastembed"

[watch]
debounce_ms = 500

[storage]
ground_dir = "~/.local/share/hallouminate/ground"