graphrag-core 0.2.0

Core portable library for GraphRAG - works on native and WASM
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
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83

# GraphRAG Configuration Template: Technical Documentation
# =========================================================
# Best for: API docs, code documentation, technical specs, architecture docs
#
# Optimized for:
# - Extracting functions, classes, modules, APIs
# - Code relationships and dependencies
# - Version and compatibility information
#
# Usage:
#   cp templates/technical.toml ./graphrag.toml
#   # or: graphrag init --template technical

output_dir = "./graphrag-technical"
approach = "algorithmic"  # Pattern-based works well for structured tech docs

# Smaller chunks to preserve code boundaries
chunk_size = 600
chunk_overlap = 100

top_k_results = 12
similarity_threshold = 0.7

[embeddings]
backend = "hash"  # Fast and works offline for tech docs
dimension = 384
fallback_to_hash = true
batch_size = 32

[entities]
min_confidence = 0.7
entity_types = [
    "FUNCTION",          # Functions, methods
    "CLASS",             # Classes, structs
    "MODULE",            # Modules, packages
    "API_ENDPOINT",      # REST endpoints, GraphQL
    "PARAMETER",         # Function parameters
    "RETURN_TYPE",       # Return types
    "ERROR_TYPE",        # Exceptions, error codes
    "VERSION",           # API versions, semver
    "DEPENDENCY",        # Libraries, crates
    "CONFIG_KEY",        # Configuration options
    "ENVIRONMENT_VAR",   # Environment variables
    "FILE_PATH",         # File references
    "URL",               # Links, endpoints
]
use_gleaning = false  # Pattern-based is often sufficient for tech docs
max_gleaning_rounds = 2

[graph]
max_connections = 20  # More connections for dependency graphs
similarity_threshold = 0.7
extract_relationships = true
relationship_confidence_threshold = 0.4

[graph.traversal]
max_depth = 5  # Deeper for module hierarchies
max_paths = 20
use_edge_weights = true
min_relationship_strength = 0.3

[retrieval]
top_k = 12
search_algorithm = "cosine"

[parallel]
enabled = true
num_threads = 0
min_batch_size = 20

[ollama]
enabled = false  # Often not needed for tech docs
host = "localhost"
port = 11434
chat_model = "llama3.2:3b"
embedding_model = "nomic-embed-text"
timeout_seconds = 30
enable_caching = true

[auto_save]
enabled = false
interval_seconds = 300
max_versions = 3