ghdigest 1.0.1

Fetch any GitHub repository into a single LLM-ready digest — metadata, directory tree, file contents, and built-in AI analysis.

RepoDigest

Fetch any GitHub repository via the GitHub API and produce an LLM-ready text digest and a structured JSON file — including the repo's "about" metadata, full directory tree, and all file contents. Then optionally pass the digest directly to an LLM for deep analysis.

---

Features

• Repo "about" metadata: description, homepage, topics, stars, forks, watchers, license, language
• Full recursive directory tree rendered as ASCII art
• All text file contents concatenated with clear separators
• Automatic filtering of binary files and noise directories
• Configurable max file size (100 KB → No limit)
• Web UI with live progress bar, results tabs, and download buttons
• LLM Analysis panel — send the digest to Groq, Mistral, OpenRouter, or Gemini with 5 preset prompts
• CLI, importable Python library, and FastAPI server
• GitHub token support (raises rate limit from 60 to 5,000 req/h)
• Outputs both .txt (LLM digest) and .json (structured data)

---

Installation

pip install -r requirements.txt

---

Quick Start

# Start the web UI
uvicorn github_ingest.server:app --reload --port 8001

# Open in browser
http://localhost:8001

---

Web UI

Visit http://localhost:8001 after starting the server.

Repo Fetch panel

• Enter any owner/repo or full GitHub URL
• Optional: GitHub token, branch, max file size (including No limit)
• Live progress bar while ingesting
• Results: about card (description, topics, stats), summary, directory tree
• Switch between TXT digest / JSON / File tree tabs
• Copy to clipboard or download .txt / .json

LLM Analysis panel (below results)

• Select a provider, paste your API key, pick a model
• Choose an analysis type or write a custom prompt
• Response streams live into the browser
• Copy the analysis to clipboard

---

LLM Providers

All providers below offer a free tier with no credit card required (except Gemini in some regions).

Provider	Free?	Context	Sign-up
Groq ⭐ recommended	Free, no card	128k tokens	console.groq.com
Mistral	Free, no card	32k tokens	console.mistral.ai
OpenRouter	Free models, no card	up to 131k	openrouter.ai/keys
Gemini	Free (may need billing)	1M tokens	aistudio.google.com/apikey

Analysis presets

Preset	What it covers
Summary	Purpose, architecture, key modules, dependencies, entry points
Architecture	Design patterns, module interactions, data flow, coupling
Security Review	Vulnerabilities, input validation, secrets, findings by severity
Onboarding Guide	Setup, structure tour, how to run, common tasks, gotchas
Code Quality	Score 1–10, organisation, docs, tests, duplication, top improvements
Custom	Write your own prompt

---

CLI Usage

# Basic — writes owner_repo.txt and owner_repo.json in current directory
python -m github_ingest owner/repo

# Full GitHub URL works too
python -m github_ingest https://github.com/owner/repo

# Authenticate (recommended — raises rate limit from 60 to 5,000 req/h)
python -m github_ingest owner/repo --token ghp_xxxxxxxxxxxx

# Or use an environment variable
export GITHUB_TOKEN=ghp_xxxxxxxxxxxx
python -m github_ingest owner/repo

# Specific branch
python -m github_ingest owner/repo --branch dev

# Custom output directory
python -m github_ingest owner/repo --output ./digests

# Limit file size (skip files larger than 100 KB)
python -m github_ingest owner/repo --max-file-size 102400

# No file size limit
python -m github_ingest owner/repo --max-file-size 0

# Only produce the JSON file
python -m github_ingest owner/repo --no-txt

# Print the text digest to stdout
python -m github_ingest owner/repo --stdout

# Suppress all progress output
python -m github_ingest owner/repo --quiet

Output files are named {owner}_{repo}.txt and {owner}_{repo}.json.

---

Python Library Usage

from github_ingest import ingest, to_txt, to_json, to_dict

# Fetch the repo
result = ingest("owner/repo", token="ghp_xxx")

# Access structured data
print(result.about["description"])
print(result.about["topics"])
print(result.about["stars"])
print(result.tree)                     # ["README.md", "src/main.py", ...]
print(result.files["README.md"])       # raw file content

# Render outputs
txt_digest  = to_txt(result)           # LLM-ready plain text
json_string = to_json(result)          # pretty-printed JSON string
data_dict   = to_dict(result)          # plain Python dict

# No file size limit
result = ingest("owner/repo", max_file_size=0)

LLM analysis from Python

from github_ingest import ingest, to_txt
from github_ingest.analyzer import analyze_stream

result = ingest("owner/repo", token="ghp_xxx")
digest = to_txt(result)

# Stream analysis using Groq (free, no card)
for chunk in analyze_stream(
    digest=digest,
    api_key="gsk_...",
    provider="groq",
    prompt_type="summary",
    model_name="llama-3.3-70b-versatile",
):
    print(chunk, end="", flush=True)

Individual fetchers

from github_ingest import fetch_repo_about, fetch_repo_tree, fetch_blob_content

about   = fetch_repo_about("owner", "repo", token="ghp_xxx")
blobs   = fetch_repo_tree("owner", "repo", sha="<commit-sha>", token="ghp_xxx")
content = fetch_blob_content("owner", "repo", blob_sha="<sha>", token="ghp_xxx")

---

FastAPI Server

Start

uvicorn github_ingest.server:app --reload --port 8001

Set GITHUB_TOKEN in the environment to apply a default token to all ingest requests:

export GITHUB_TOKEN=ghp_xxxxxxxxxxxx
uvicorn github_ingest.server:app --reload --port 8001

Endpoints

Method	Path	Description
GET	/	Web UI
GET	/health	Liveness check
POST	/ingest	Ingest repo → JSON
POST	/ingest/txt	Ingest repo → plain-text digest
GET	/ingest?repo=owner/repo&fmt=json	Ingest via query params
GET	/analyze/providers	List providers, models, prompt types
POST	/analyze	Analyze digest → SSE stream
POST	/analyze/test	Test an API key with a minimal request

POST /ingest

{
  "repo": "owner/repo",
  "token": "ghp_xxxxxxxxxxxx",
  "branch": "main",
  "max_file_size": 524288
}

Set max_file_size to 0 for no limit. token falls back to the GITHUB_TOKEN env var.

POST /analyze

{
  "digest": "<your .txt digest>",
  "api_key": "gsk_...",
  "provider": "groq",
  "prompt_type": "summary",
  "model": "llama-3.3-70b-versatile"
}

provider options: groq · mistral · openrouter · gemini
prompt_type options: summary · architecture · security · onboarding · quality · custom
Returns a Server-Sent Events stream: data: {"text": "..."} chunks, ending with data: [DONE].

POST /analyze/test

{ "api_key": "gsk_...", "provider": "groq", "model": "llama-3.3-70b-versatile" }

Sends a minimal request to verify the key works before sending the full digest.

Interactive API docs

http://localhost:8001/docs

---

Output Format

.txt digest

================================================================
REPOSITORY: owner/repo  [main]
================================================================

ABOUT
-----
Description : A great project
Homepage    : https://example.com
Language    : Python
License     : MIT
Topics      : python, llm, tools
Stats       : Stars: 1,234 | Forks: 56 | Watchers: 89 | Open issues: 12

================================================================
SUMMARY
================================================================
Files ingested : 42
Files skipped  : 8

================================================================
DIRECTORY STRUCTURE
================================================================
repo/
├── README.md
├── src/
│   ├── main.py
│   └── utils.py
└── tests/
    └── test_main.py

================================================================
FILES
================================================================
──── README.md ────────────────────────────────────────────────
# My Project
...

.json output

{
  "repository": "owner/repo",
  "branch": "main",
  "about": {
    "description": "A great project",
    "topics": ["python", "llm"],
    "stars": 1234,
    "license": "MIT"
  },
  "summary": { "files_ingested": 42, "files_skipped": 8 },
  "tree": ["README.md", "src/main.py"],
  "files": { "README.md": "# My Project\n..." },
  "skipped": ["assets/logo.png"]
}

---

GitHub Rate Limits

	Unauthenticated	With token
Limit	60 req/hour	5,000 req/hour
Small repo (~20 files)	Works	Works
Medium repo (~200 files)	Hits limit	Works
Large repo (1000+ files)	Fails	Works

Get a free token at github.com/settings/tokens/new — no scopes needed for public repos.

---

Filtered by Default

Binary extensions: images, audio, video, archives, compiled objects, fonts, office docs, databases

Noise directories: .git, node_modules, __pycache__, .venv, venv, dist, build, .next, .pytest_cache, and more

Files over 500 KB are skipped by default. Override with --max-file-size <bytes> or 0 for no limit.

---

License

MIT License. See the LICENSE file for details.