Skip to main content

RepoMemory

AI-powered code retrieval engine — index any GitHub repo, search with natural language

PyPI PyPI Downloads Python 3.11+ License: MIT CI

Point it at any GitHub URL. Get token-budget-aware context packs — ready to paste into any LLM. Free to run, free to deploy.


What is RepoMemory?

When you ask an LLM to fix a bug or trace a feature, it needs the right source files. Pasting the whole codebase wastes the context window. Guessing which files to include misses critical pieces.

RepoMemory solves this with a hybrid retrieval pipeline that runs on any public or private GitHub repo:

repomemory index https://github.com/pallets/flask
repomemory search "Where is request routing handled?"
Query  →  Task Classification (trace_flow)
       →  BM25 Lexical + FAISS Semantic + Fuzzy Path + Symbol search  (parallel)
       →  Reciprocal Rank Fusion  →  Top-20 ranked files
       →  Dependency-graph expansion  →  Re-rank with adaptive weights
       →  Token-budget packer  →  Context pack ready to paste into any LLM
       →  (optional) Groq AI summary

Install

# Core — CLI + library, uses HuggingFace API for embeddings (no GPU needed)
pip install repomemory

# With local embeddings (~80 MB model download, fully offline)
pip install "repomemory[local]"

# With FastAPI web server
pip install "repomemory[server]"

# With Groq LLM explanations (free API key)
pip install "repomemory[llm]"

# Everything
pip install "repomemory[all]"

Quick Start

CLI

# Index a repo
repomemory index https://github.com/pallets/flask

# Search it
repomemory search "How does request routing work?"

# With AI explanations (free Groq key)
export REPOMEMORY_GROQ_API_KEY=gsk_...
repomemory search "Where is token rotation handled?"

# Adjust result count and token budget
repomemory search "auth middleware" --top-k 10 --budget 4000

# Force a task mode
repomemory search "test coverage for auth" --mode test_lookup

# Private repos
repomemory index https://github.com/myorg/private-repo --token ghp_...

# List indexed repos
repomemory list

# Start the web UI + API server
repomemory serve

Python library

from repomemory import RepoMemory

rm = RepoMemory()

# Index
repo = rm.index("https://github.com/pallets/flask")

# Search
result = rm.search("How does request dispatching work?")
for file in result.context_pack.files:
    print(f"{file.path}  score={file.relevance_score:.2f}")
    print(file.snippets[0].content[:300])

REST API

# Start the server
pip install "repomemory[server]"
repomemory serve

# Index a repo
curl -X POST http://localhost:8000/api/repos \
  -H "Content-Type: application/json" \
  -d '{"url": "https://github.com/pallets/flask"}'

# Search
curl -X POST http://localhost:8000/api/search \
  -H "Content-Type: application/json" \
  -d '{"repo_id": 1, "query": "How does routing work?", "token_budget": 8000}'

Features

Feature Details
Hybrid search BM25 lexical + FAISS semantic (384-dim) + fuzzy path + symbol name, fused with Reciprocal Rank Fusion
Dependency-graph retrieval Builds file-level import edges at index time; expands results through related files via BFS
Adaptive weight learning Online SGD learning from user feedback (accept / dismiss / thumbs); falls back to static mode weights
Symbol-aware indexing tree-sitter extracts functions, classes, and methods from Python, JavaScript, and TypeScript
5 Task Modes bug_fix, trace_flow, test_lookup, config_lookup, general — auto-detected from query or set manually
Token-budget packer Greedy packer respects any token limit (default 8 000 tokens, configurable up to 100 000+)
Behavioral memory Frecency scoring from opened / accepted / thumbs-up actions; boosts relevant files in future queries
RAG evaluation End-to-end pipeline scoring retrieval impact on LLM answer quality (relevance, completeness, faithfulness)
Flexible embeddings Local sentence-transformers (offline) or free HuggingFace Inference API
AI explanations Optional Groq LLM (free tier) explains why each result matters
Incremental indexing SHA-256 per file; only changed files are re-embedded on re-index
Web UI + REST API React 19 frontend + FastAPI backend; deploy on Render + Vercel (both free)
Export as Markdown Copy context pack as a formatted Markdown block to paste directly into an LLM prompt

Task Modes

RepoMemory classifies each query and adjusts retrieval weights automatically:

Mode Auto-detected from What it boosts
bug_fix error, exception, crash, fix, traceback Lexical signal, error-adjacent files
trace_flow trace, flow, route, handler, how does...work Symbol matching, call-chain ordering
test_lookup test, spec, mock, fixture, coverage Path matching for tests/ / spec/ dirs
config_lookup config, env, setting, yaml, toml Path matching for config-like files
general (fallback) Balanced across all signals

Configuration

All settings use the REPOMEMORY_ env prefix (powered by pydantic-settings):

export REPOMEMORY_HF_API_KEY=hf_...          # HuggingFace free API key
export REPOMEMORY_GROQ_API_KEY=gsk_...       # Groq free API key (for AI summaries)
export REPOMEMORY_EMBEDDING_PROVIDER=local   # 'local' or 'huggingface'
export REPOMEMORY_DATA_DIR=/data/repomemory  # where SQLite + FAISS live
export REPOMEMORY_TOKEN_BUDGET=16000         # default context pack size


License

MIT © Aayush Kumar

Metadata

Release files for repomemory 0.2.3

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for repomemory 0.2.3
File Size Uploaded Provenance Actions
repomemory-0.2.3.tar.gz 55.0 kB With provenance Details for repomemory-0.2.3.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for repomemory 0.2.3
File Interpreter ABI Platform Actions
repomemory-0.2.3-py3-none-any.whl Python 3 none any Details for repomemory-0.2.3-py3-none-any.whl

Total release size: 113.7 kB

Release files / repomemory-0.2.3.tar.gz

Download URL repomemory-0.2.3.tar.gz
Size 55.0 kB
Tags Source
SHA-256 checksum
How to use checksums
b865c0db82747e700c14278b15b82bac6934c925f85d31ad11cab1a21eaaf730
BLAKE2b-256 checksum
How to use checksums
7a5dd1c81d79c5a3f4b104437f71bb1620aaad45cb5c39aa1e171ef5d750e58f
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/6.1.0 CPython/3.13.12

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Apr 17, 2026.

Transparency log

Release files / repomemory-0.2.3-py3-none-any.whl

Download URL repomemory-0.2.3-py3-none-any.whl
Size 58.7 kB
Tags Python 3
SHA-256 checksum
How to use checksums
83139ab6ad2461d10c8c77448c8d60bd5af43d7b974ce41fc9ec747b98f916ac
BLAKE2b-256 checksum
How to use checksums
da4d7327f1fa9fa09d8ec4a6a1cdafc7e62d672cd119ac81ec198c01d79a26f6
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/6.1.0 CPython/3.13.12

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Apr 17, 2026.

Transparency log

Security for repomemory 0.2.3

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for repomemory-0.2.3.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Apr 17, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page