Configuration

GitHelp uses YAML configuration files stored in configs/, plus project-specific configuration files generated under data/projects/.

There are two levels of configuration:

configs/
→ global GitHelp settings

data/projects/<project_name>/
→ generated configuration and corpus for a selected target project

app_config.yaml

This file stores application-level settings.

Example:

app_title: GitHelp
app_subtitle: Ask questions about a project's documentation
show_sources: true
default_top_k: 5

project_profile: mmore

llm:
  provider: qwen
  model_name: Qwen/Qwen3-4B
  max_new_tokens: 512
  temperature: 0.0
  enable_thinking: false

project:
  config_path: configs/project_config.yaml

MMORE sparse indexing currently requires Transformers 4.x. GitHelp pins:

transformers>=4.51.0,<5

If MMORE index building fails with a tokenizer error such as BertTokenizer has no attribute batch_encode_plus, reinstall the compatible version:

python -m pip install "transformers>=4.51.0,<5"

Main fields

Field	Meaning
app_title	Parsed application title. The current Streamlit header still renders GitHelp directly.
app_subtitle	Parsed application subtitle. The current Streamlit header still renders its caption directly.
show_sources	Parsed display default; persisted Streamlit state takes precedence in the current UI.
default_top_k	Parsed retrieval default; persisted Streamlit state currently initializes the sidebar value.
project_profile	Project-specific behavior profile, for example generic or mmore.
project.config_path	Project config used to identify the indexed project when preparing prompts. This is not changed automatically when Streamlit builds another project.
llm.provider	LLM provider used by GitHelp.
llm.model_name	Model name used by the provider.
llm.max_new_tokens	Maximum number of generated tokens.
llm.temperature	Generation temperature.
llm.enable_thinking	Whether to enable Qwen thinking mode.

project_config.yaml

This file describes the target project being indexed.

The default configuration is:

configs/project_config.yaml

When using the Streamlit interface, GitHelp also generates a project-specific config:

data/projects/<project_name>/project_config.yaml

Example for MMORE:

project_name: mmore
package_name: mmore

repo_path: /absolute/path/to/mmore
docs_path: /absolute/path/to/mmore/docs/source
code_path: /absolute/path/to/mmore/src/mmore

include_yaml_configs: true
yaml_config_paths:
  - /absolute/path/to/mmore/examples
  - /absolute/path/to/mmore/production-config

include_repo_structure: true
repo_structure_max_depth: 4

Main fields

Field	Meaning
project_name	Project name stored in metadata and used for project folders.
package_name	Python package prefix used to build full module names.
repo_path	Root folder of the target repository.
docs_path	Folder containing Markdown or reStructuredText docs.
code_path	Python source folder used for docstring extraction.
include_yaml_configs	Whether to include YAML files in the corpus.
yaml_config_paths	Folders scanned for .yaml and .yml files.
include_repo_structure	Whether to add a synthetic repository tree document.
repo_structure_max_depth	Maximum depth of the generated repository tree.

Generated project configs also contain include_extensions and exclude_patterns. These values are parsed and retained for future filtering, but the current corpus builder selects sources through the configured paths, source-specific extensions, and loader-level exclusions instead.

indexing_config.yaml

This file contains prototype corpus/retrieval settings and the defaults used by scripts/build_index.py.

include_markdown: true
include_code_docstrings: true
include_signatures: true
include_code_snippets: false

chunk_size: 1200
chunk_overlap: 150

top_k: 5
retrieval_backend: simple
collection_name: mmore_docs
mmore_index_config_path: configs/mmore_index_config.yaml

The current index command reads collection_name and mmore_index_config_path. Fields describing source inclusion, chunking, top_k, and retrieval_backend are currently reserved configuration: the corpus builder and Streamlit sidebar do not consume them. Corpus construction instead relies on Markdown sections and extracted code documentation records.

mmore_index_config.yaml

This file is passed to MMORE when building the index.

indexer:
  dense_model:
    model_name: sentence-transformers/all-MiniLM-L6-v2
    is_multimodal: false
  sparse_model:
    model_name: splade
    is_multimodal: false
  db:
    uri: ./data/indexes/mmore/githelp.db
    name: my_db

collection_name: mmore_docs
documents_path: data/processed/mmore_corpus.jsonl

mmore_retriever_config.yaml

This file configures MMORE retrieval.

db:
  uri: ./data/indexes/mmore/githelp.db
  name: my_db

hybrid_search_weight: 0.5
k: 5
collection_name: mmore_docs
use_web: false
reranker_model_name: null

Both MMORE config files currently point to one local Milvus Lite database and the shared mmore_docs collection. Project-specific corpora and exports are isolated under data/projects/, but native MMORE indexes are not yet isolated per project.

data/app_state.json

The Streamlit app persists the latest local UI state in:

data/app_state.json

It can contain:

{
  "project_name": "mmore",
  "project_path": "/path/to/mmore",
  "corpus_path": "/path/to/githelp/data/projects/mmore/corpus.jsonl",
  "project_config_path": "/path/to/githelp/data/projects/mmore/project_config.yaml",
  "mmore_corpus_path": "/path/to/githelp/data/projects/mmore/mmore_corpus.jsonl",
  "collection_name": "mmore_docs",
  "indexing_mode": "mmore",
  "backend": "simple",
  "top_k": 5,
  "use_llm": true,
  "show_sources": true,
  "show_full_sources": false,
  "show_debug": false
}

This file is machine-specific and should normally be ignored by Git.