Erold is a context engine for AI coding agents. It gives your AI agent persistent memory that survives across sessions using Smart Strip compression, which delivers 2-3x more context in the same token window. With just 3 MCP tools and passive capture hooks, agents spend time coding instead of bookkeeping.

What is Smart Strip compression?

Smart Strip is a lossless fact-extraction technique that compresses coding session transcripts by 2-3x. It classifies facts as self-identifying (bare values like file paths, error codes) or context-dependent (values needing minimal disambiguation), then strips all filler. The compressed format survives unlimited re-compression cycles without drift.

Yes. The full context engine -- Smart Strip compression, 3 MCP tools, passive hooks, and 158+ coding guidelines -- is free forever. Bring your own AI API key. Paid plans add team shared memory and the RPA privacy anonymization layer.

What AI tools does Erold support?

Erold works with any MCP-compatible AI tool, including Claude Code, Claude Desktop, Cursor, Windsurf, and custom AI agents. 3 MCP tools, one configuration command, instant persistent memory.

Why only 3 MCP tools?

More tools means more tokens spent on tool selection and more chances for the agent to pick the wrong one. Erold exposes three focused tools -- get_context, save_memory, and log_work -- while passive hooks handle the rest automatically.

What does re-compression resistant mean?

When compressed memory is loaded, used, and re-saved, most compression formats lose facts or introduce drift. Smart Strip encoding produces stable output through unlimited compression cycles -- the same facts survive every round without decay or hallucination.

What is the RPA privacy layer?

The Redaction and Privacy Anonymization layer strips proprietary code, names, and sensitive details before context is sent to external LLMs. Your compressed memory stays useful for the agent while keeping sensitive information local. Available in the Enterprise plan.

What integrations does Erold support?

Erold works with Claude Code, Cursor, Windsurf, and Claude Desktop via the MCP Server. Any MCP-compatible AI tool can connect to Erold with a single configuration command.

Introduction to Erold

Erold is a context engine for AI coding agents. It gives your agent persistent, compressed memory that survives across sessions -- powered by Smart Strip compression that delivers 2-3x more context in the same token window.

The Problem

AI coding agents are powerful but amnesiac. Every new session starts from zero:

Agents re-discover decisions they already made
Failed approaches get repeated because they were forgotten
Context windows fill up with redundant, verbose summaries
Re-summarized memory drifts from the original facts over time

Erold solves this with Smart Strip compression and passive capture hooks, delivered through just 3 MCP tools.

Key Features

Smart Strip Compression

Lossless fact extraction that compresses coding session transcripts by 2-3x. Every discrete fact is classified as self-identifying (bare values like file paths, error codes, commands) or context-dependent (values needing one disambiguating word). All filler is stripped. The result survives unlimited re-compression cycles without drift.

3 MCP Tools

Agents spend time coding, not bookkeeping. Three focused tools replace dozens of granular ones: get_context loads compressed memory, save_memory persists new knowledge, and log_work records completed work. Works with Claude Code, Cursor, Windsurf, and any MCP-compatible tool.

Passive Capture Hooks

Hooks observe your agent's work silently -- session transcripts, file changes, decisions, failed approaches. Memory builds itself without you or your agent explicitly saving anything. No workflow interruptions, no manual updates.

158+ Coding Guidelines

Lean, AI-optimized coding guidelines as structured data. Sharp rules that agents apply instantly. Covering React, Next.js, FastAPI, security, and more. Community-contributed and always growing.

RPA Privacy Layer Enterprise

Anonymize code, names, and proprietary details before sending context to external LLMs. Keep sensitive information local while your agent gets the context it needs.

How It Works

Connect your agent - Add Erold as an MCP server (one command, under 2 minutes)
Agent works, Erold remembers - Hooks passively capture work and compress it via Smart Strip
Next session starts smart - Agent loads compressed context automatically, knows what was tried and what worked

Next Steps

Quick Start Guide - Get persistent memory working in 2 minutes
Core Concepts - Understand Smart Strip, hooks, and memory
MCP Setup - Configure your AI agent integration