repodanta

repository analysis tool for developers who need to understand large codebases fast.

no docs needed. just point it at a repo.

---

what it does

repodanta combines structural analysis, dependency graphing, and ai reasoning to answer questions about any codebase.

tested on fastapi — 1118 modules, 107k lines of code, analyzed in seconds.

---

commands

inspect

get a structural overview of any repo.

repodanta inspect /path/to/repo

repository overview
----------------------------------------
modules           : 1118
functions         : 4390
lines of code     : 107283
avg funcs/module  : 3.93
----------------------------------------
largest module    : tests.test_include_router_defaults_overrides.py (7304 LOC)
largest function  : tests.test_include_router_defaults_overrides.py:test_openapi (6865 lines)
----------------------------------------
dependency hotspots
  fastapi.testclient.py (fan-in=449, fan-out=1, score=449)
  fastapi.responses.py  (fan-in=74,  fan-out=3, score=222)
  fastapi.routing.py    (fan-in=15,  fan-out=12, score=180)

orchestration modules
  fastapi.applications.py (fan-out=14)
  fastapi.routing.py      (fan-out=12)

core modules
  fastapi.testclient.py (fan-in=449)
  fastapi.responses.py  (fan-in=74)

risky modules
  fastapi.testclient.py (risk=899)
  fastapi.responses.py  (risk=151)
----------------------------------------

fastapi's testclient.py has a fan-in of 449 — nearly half the codebase depends on it. you'd never know that just by reading the code.

---

trace

find how a function is reached in the system.

repodanta trace /path/to/repo get_request_handler

call chain
----------------------------------------
fastapi.routing.py:get_route_handler
?:get_request_handler
----------------------------------------

---

query

ask questions about the codebase in plain english.

repodanta query /path/to/repo "how does request routing work"

request routing in fastapi flows through routing.py.
get_route_handler determines which endpoint matches
the incoming request and returns the handler function.

---

install

pip install repodanta
repodanta --help

if the repodanta command is not found after install, add pip's bin folder to your PATH:

echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.bashrc
source ~/.bashrc

requirements

• ollama running locally with qwen2.5

ollama pull qwen2.5
ollama serve

on first run, sentence-transformers will download the embedding model automatically. you may see an unauthenticated warning from huggingface — this is harmless.

gpu support

repodanta's inference runs through ollama. gpu acceleration depends on ollama's hardware compatibility. check ollama's documentation for supported gpus. amd gpu support is limited by ollama, not repodanta.

how it works

structural analysis    →  dependency graph, fan-in/fan-out, hotspot detection
call chain tracing     →  ast-based function call graph traversal
semantic search        →  embeddings + faiss + ai reasoning over retrieved chunks

repodanta doesn't just describe code. it understands how the system is connected.

---

built with

• python ast — static analysis
• sentence-transformers — code embeddings
• faiss — vector search
• ollama (qwen2.5) — local ai reasoning, no api key needed

---

status

early stage. currently supports python repos only.

built by Yogadharshan.