Quickstart

This page explains the shortest path to run GitHelp locally.

GitHelp requires Python 3.10 or higher.

GitHelp can be used in two ways:

  • through the deployed EPFL interface, available from the EPFL network or VPN;

  • locally, by cloning the repository and running the Streamlit app.

The local Streamlit interface is the recommended way to test GitHelp during development. The command-line scripts remain available for debugging and development.

1. Clone the repository

git clone https://github.com/EPFLiGHT/GitHelp.git
cd GitHelp

All commands below must be run from the root of the GitHelp repository.

2. Create an environment

conda create -n githelp python=3.10 -y
conda activate githelp

3. Install the project

python -m pip install -e .

This installs Streamlit, the local Qwen dependencies, and MMORE. The native MMORE backend is heavier than the simple backend and may download models during its first indexing or retrieval run.

4. Launch the Streamlit app

streamlit run app/streamlit_app.py

The interface opens locally, usually at:

http://localhost:8501/

The deployed EPFL interface is separate from this local run and is available from the EPFL network or VPN at:

http://gpu217.rcp.epfl.ch:1312/githelp/

5. Select a project

In the Streamlit interface, use the Project setup section.

Enter the local path to the repository. For example, to query questions on MMORE:

/Users/<user>/path/to/mmore

or select the GitHub repository option and enter:

https://github.com/swiss-ai/mmore

For GitHub URLs, GitHelp clones the repository into data/repositories/ and then runs the same corpus, indexing, retrieval, and RAG pipeline on that local copy.

6. Prepare the project

Click one of the project build buttons:

Build simple index
Build MMORE index

GitHelp generates a dedicated project folder:

data/projects/<project_name>/

For MMORE, both modes create for example:

data/projects/mmore/project_config.yaml
data/projects/mmore/corpus.jsonl

The generated corpus can include:

  • Markdown and reStructuredText documentation;

  • Python module, class, function, and method docstrings, plus function and method signatures extracted with ast;

  • YAML configuration files;

  • a synthetic repository structure document.

7. Ask questions

After the corpus is built, use the chat input at the bottom of the Conversation section. The input remains disabled until a valid project corpus is available.

Choose an indexing mode

GitHelp supports two indexing modes.

Simple index

The simple index builds a GitHelp JSONL corpus and uses the local simple retriever.

Use it when:

  • you want a quick setup;

  • you are debugging corpus extraction;

  • MMORE is not installed or not configured.

MMORE index

The MMORE index is the main semantic retrieval mode. It builds the GitHelp corpus, exports it to MMORE format, and builds the MMORE index.

Use it when:

  • MMORE is installed;

  • you want to use the main retrieval backend;

  • the project is ready to be indexed.

After building the MMORE index, select:

Retrieval backend: mmore

The simple backend remains the recommended first run for debugging or quick corpus checks.

Example questions:

How do I configure indexing?
Which Milvus parameters are used in the ColPali config?
Where are the example configs located?

8. Inspect sources

By default, GitHelp displays retrieved sources under the answer.

The sidebar options let you:

  • show or hide retrieved sources;

  • show full source content;

  • show debug information;

  • switch between simple and mmore retrieval;

  • enable or disable LLM generation.

When the mmore backend is selected, the diagnostics distinguish native index retrieval from the lexical corpus fallback:

native_index
corpus_fallback

The fallback searches the exported mmore_corpus.jsonl lexically. It does not use native MMORE/Milvus vector retrieval.

9. Persistent app state

GitHelp stores the last selected project and UI settings in:

data/app_state.json

This allows the interface to restore the previous project, corpus path, backend, and display options after closing and reopening Streamlit.

This file is local state and should normally not be committed.

10. Optional: command-line corpus build

The default command still works:

python scripts/build_corpus.py

It reads:

configs/project_config.yaml

and writes:

data/processed/corpus.jsonl

You can also build a project-specific corpus manually:

python scripts/build_corpus.py \
  --config data/projects/mmore/project_config.yaml \
  --output-path data/projects/mmore/corpus.jsonl

11. Optional: MMORE indexing

The mmore backend retrieves from an MMORE index. Building a corpus alone is not enough to update that index.

For a full MMORE-backed workflow, the steps are:

build_corpus.py
→ export_mmore_corpus.py
→ build_index.py
→ ask with backend mmore

For local development, the simple backend is useful when you want to inspect the GitHelp corpus before exporting and indexing it with MMORE.

Note

Project corpora and MMORE export files are stored separately. The default app configuration nevertheless uses the MMORE-specific profile, and native MMORE indexing currently uses one shared mmore_docs collection. Use project_profile: generic for another project and rebuild the native index when switching its indexed corpus.