skegdb

skegdb / skeg

Public

Vector database for AI agents. Context layer, multi-tenant, RAM-frugal. Runs anywhere your model runs.

10
0
94% credibility
Found May 30, 2026 at 10 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Rust
AI Summary

skeg is a vector database and context storage system designed specifically for personal AI setups where a language model already occupies most of the computer's memory. Unlike other vector databases that expect to own all available RAM, skeg stores most data on the SSD while keeping a small, fast cache in memory. It provides both traditional key-value storage and vector similarity search in a single database, speaks the Redis wire protocol so existing tools work with it, and is optimized to run efficiently alongside large language models on Apple Silicon and ARM-based machines. The project is open source under Apache 2.0 with well-documented performance benchmarks and comprehensive test coverage.

How It Works

1
💡 Discovering skeg

You learn about skeg when researching vector databases for your local AI setup. You notice it uses 10x less memory than other options.

2
📦 Installing skeg

You install skeg with one command on your Mac or Linux machine. It runs locally alongside your language model.

3
⚡ Everything stays fast

Your language model keeps using the memory it needs, while skeg quietly stores vectors and data on your fast SSD. Queries still return in milliseconds.

4
🔌 Connecting your tools

You use your existing Redis tools and libraries to talk to skeg. It understands the same commands your favorite database tools speak.

5
Choose your path
🔤
Key-value data

Store and retrieve text, numbers, and settings just like a regular database

🔍
Vector search

Create an index of embeddings and find similar items by asking questions

6
🌐 Running your AI app

Your application sends requests to skeg, which retrieves relevant context to give your AI model the information it needs to answer well.

✨ Everything works together

Your AI assistant has fast access to both your conversation history and related knowledge from your vector index, all while staying within your memory budget.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 10 to 10 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 skeg?

Skeg is a vector database and context layer for AI agents, written in Rust. It stores both vectors and key-value pairs in a single engine, designed to run on the same machine as your language model. The standout claim: it uses roughly 400MB of RAM for 1 million vectors at high recall, versus 4+ GB for comparable systems. It speaks two protocols -- a native binary format on port 7379 and Redis-compatible RESP3 on port 6379, so you can query it with standard Redis tooling or a native client.

Why is it gaining traction?

The hook is memory efficiency on resource-constrained hardware. Most vector databases assume they own the machine; skeg assumes the model already does. It keeps the index on SSD with a small bounded cache, using multiple quantization tiers (int8, product quantization, and TurboQuant at 1-4 bits per coordinate) to trade precision for memory. The Vamana graph is walked on disk with hot-page caching. Benchmarks on M1 show 489 queries per second at 3.6ms p99 latency while staying under 420MB RSS. For developers running local models or working with limited RAM budgets, this is the only open-source option purpose-built for that constraint.

Who should use this?

Developers running local AI inference -- particularly on Apple Silicon or ARM servers -- who want vector search without starving their model of memory. RAG application developers building personal AI tools where a 32GB M-series chip needs to run both the model and the vector store. Teams evaluating vector database options for RAG pipelines and wanting to compare raw memory usage. Early adopters comfortable with a v0.1.0 release who want to follow a project from the ground floor.

Verdict

Skeg solves a real problem for a specific niche: low-RAM vector search alongside a loaded model. The Redis compatibility and simple install (Homebrew, Docker, or cargo) lower the barrier to trying it. At 10 stars and v0.1.0, the credibility score is 0.949999988079071% -- this is very early software. Docs exist but are thin, and native Linux validation is pending. Worth evaluating for the memory-constrained use case, but treat it as an experimental project to watch rather than a production-ready choice.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.