ContextOS

Memory infrastructure for AI agents.

Give AI applications persistent, searchable, intent-aware context across millions of files without sending entire repositories to the LLM.

Documentation View on GitHub

$npm i -g @siddharthakatiyar/contextos && contextos init

The Problem

AI agents break when context windows overflow. ContextOS replaces naive file loading with an intelligent retrieval pipeline.

Without ContextOS

read auth.ts

read database.ts

read queue.ts

read redis.ts

...

Total Tokens

Context Window Warning

Hallucination

With ContextOS

Expand Graph

Rank Symbols

Extract

Total Tokens

1,328

Correct Answer

See it in action

A side-by-side comparison of an AI Agent answering a question about the repository.

Without ContextOS

42K+ Tokens

Spawns expensive Explore subagents, blindly greps files, and suffers from massive context bloat.

With ContextOS

293 Tokens

Instant semantic retrieval, pinpoint accuracy, and minimal context usage in a single tool call.

Retrieval Pipeline

Watch how a repository transforms into a surgical context package.

Repository

Index

Graph Expansion

Intent Ranking

Compression

LLM Context

Pipeline Stage 1

Repository

Your raw codebase across thousands of files.

Graph Expansion

Search finds keywords. Graph Expansion finds reality. Watch how ContextOS automatically pulls in related dependencies to build a complete context package.

Waiting for query...

indexer.ts

cli.ts

api.ts

database.ts

schema.ts

migrations.ts

Upto0%

Context token reduction

312ms

Average retrieval latency

SQLite

Backed, zero external DBs

AST

Aware semantic chunking

Ready to build?

macOS / Linux / Windows • Zero configuration required