shreyash-sharma

shreyash-sharma / provenant

Public

Wiki-based retrieval for AI coding agents. 65× token reduction, +24pp Coverage@5 on SWE-bench Verified.

www.shreyashsharma.comwritingprovenant ai codebase-indexing developer-tools llm mcp
12
8
89% credibility
Found May 29, 2026 at 20 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Provenant is a local codebase intelligence tool that gives AI coding assistants a memory layer - parsing code, building dependency graphs, generating wiki pages, and exposing everything through MCP tools so agents can answer questions about your codebase with citations instead of guessing blindly.

How It Works

1
💡 Discovering a smarter coding assistant

You hear about Provenant - a tool that gives your AI coding assistant a memory layer so it understands your codebase instead of guessing blindly.

2
📦 Installing Provenant

You install Provenant with a simple command. It works locally on your machine, keeping your code private.

3
🔍 Indexing your repository

You point Provenant at your project folder. It reads through all your files, maps how they connect, and studies your git history.

4
✨ Generating a codebase wiki

Provenant creates plain-English explanations for every file and module in your project - automatically, using AI.

5
Connecting to your favorite editor
🎯
Claude Code

Provenant becomes your pair programmer's memory - it knows your codebase inside and out

💻
Cursor or Windsurf

Your AI assistant gains deep context about your project structure and dependencies

🤖
Any MCP-compatible tool

Provenant works with any AI tool that speaks the MCP protocol

6
💬 Asking questions about your code

You ask your AI assistant about architecture, ownership, or where code lives. It answers with citations pointing back to your actual files.

🎉 Your AI assistant becomes an expert

Instead of re-reading thousands of lines of code, your assistant retrieves focused, cited context. Changes are safer, answers are grounded, and everything just works.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 20 to 12 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is provenant?

Provenant is a Python CLI and MCP server that gives AI coding agents a persistent memory layer for your codebase. Instead of agents repeatedly opening raw files and wasting context tokens, Provenant indexes your repository into searchable wiki pages, dependency graphs, and risk signals before the agent starts working. It runs locally, stores everything in your repo's `.provenant/` directory, and exposes eight MCP tools for retrieval, risk analysis, dead code detection, and git archaeology.

Why is it gaining traction?

The hook is simple: agents are expensive, and naive file loading burns through context tokens fast. Provenant claims 60-65x token reduction on Flask/Django workloads and +7.6 percentage points improvement in file localization on SWE-bench Verified. It also self-heals—low-confidence answers trigger background wiki rewrites, so the index gets smarter over time. The MCP compatibility means you get the same memory layer across Claude Code, Cursor, Windsurf, Cline, and Copilot without vendor lock-in.

Who should use this?

Developers working with large or legacy codebases where "why does this file exist?" matters. Teams running AI coding agents who want cited, grounded answers instead of free-floating summaries. Anyone whose agent keeps opening the wrong files before making progress. If you're doing SWE-bench-style automation or maintaining a repo with hidden dependencies and architectural debt, this is worth a look.

Verdict

Provenant solves a real problem with a thoughtful architecture, and the self-healing retrieval is a genuinely clever idea. However, the credibility score sits at 0.9% with only 12 stars—early-stage project, limited community validation, and the benchmarks come from the author. Try it on a side project before betting your team on it. The token savings claim is compelling enough to justify a weekend experiment.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.