Open source · Local-first · MIT License

Your AI Forgot Your
Project Again.
We Fixed That.

Without Eidos1,245 words
With Eidos57 words

Eidos Memory gives any AI CLI persistent, compressed context. 95.6% fewer tokens. Works with Claude, Gemini, Qwen, and 9+ others.

Get Started
GitHub starsnpm downloads106 tests passingMIT License
Eidos Dashboard — localhost:7842

Eidos Dashboard

Memory engine active

Memory Active

Token Savings

95.6%

Context Chunks

2,847

+42 indexed today

Active Projects

6

3 synced via CRDT

Knowledge Graph

Recent Activity

Injected context

claude → auth module

2s ago

Indexed

+42 chunks from ./src

14s ago

Compressed

4,200 → 183 tokens

1m ago

Retrieved

7 vectors for query

3m ago

Synced

CRDT merge complete

5m ago
eidos-demo.gif
Eidos Memory demo — AI memory engine saves 96% tokens
Benchmark

Token Savings Benchmark

Real numbers from actual usage. Run eidos demo to see them yourself.

Debugging a bug:1,245 words57 wordsThe AI already knew.
Scenario
Without Eidos Memory
With Eidos Memory
Savings
95.6%
At 50 prompts/day, save ~$90/mo per developer
Features

Problems We Solve

One install, zero config. Eidos Memory handles context for every AI tool on your machine.

Universal

Your AI tools each have their own memory — or none at all.

Eidos injects context into Claude, Gemini, Qwen, Aider, and 9+ other CLIs. One memory for every tool.

Local-first

Your code should never leave your machine.

Embeddings, AST parsing, and graph storage all run locally in ~/.eidos/. Zero cloud dependencies.

Privacy Firewall

Accidentally sending API keys or secrets to your AI?

.eidosignore excludes sensitive files. Automatic secret redaction for API keys, JWTs, and env vars.

Adaptive Token Budget

Not every question needs the whole codebase.

Eidos auto-detects whether you're debugging, implementing, or recalling — and adjusts context size accordingly.

Vector Search

Keyword search misses what matters. You need semantic understanding.

384-dim MiniLM or optional 768-dim BGE embeddings with sqlite-vec ANN for fast similarity search.

CRDT Sync

Your team's AI memory should sync — securely.

AES-256-GCM encrypted team sync via shared folder or self-hosted HTTP relay. CRDT-based conflict resolution.

Web Dashboard

See exactly what your AI knows — and what it's saving you.

vis-network graph explorer + savings charts at localhost:7842. Visualize your knowledge graph in real-time.

MCP Server

Your AI should be able to query its own memory.

Full MCP server with search, assemble, remember, and feedback tools for Claude Desktop, Cursor, and Continue.

Sub-10ms Latency

Context retrieval shouldn't slow you down.

Under 10ms latency. No noticeable delay in your AI coding workflow.

Workflow

How It Works

Three steps. Zero config. Your AI gets persistent memory automatically.

1

Index

Eidos scans your codebase, chunks it into semantic units (functions, classes, modules), and creates vector embeddings. Everything stays local in ~/.eidos/.

2

Retrieve

When you ask a question, Eidos finds the most relevant chunks via vector search. Not the whole file — just the functions, decisions, and context that matter.

3

Compress

The retrieved context is compressed: function bodies become skeletons, conversation turns become micro-summaries, diffs become patch-only. 90-98% reduction.

Comparison

Why EidosCore?

EidosCore replaces claude-mem, memory, graphify, and caveman with one tool.

Featureclaude-memmemorygraphifyEidosCore
Code-aware semantic search
Conversation memory
Automatic context injection
Token-budgeted assembly
Universal CLI adapters (9+)
Session continuity (QMS)
Self-learning retrieval
Privacy-first (local-only)
MCP server (14 tools)
CRDT team sync

Last updated May 2026. Based on publicly available feature lists.

Audience

Is Eidos Memory For You?

Eidos Memory is built for developers who:

Use Claude Code, Gemini CLI, Qwen, or Aider daily

Waste tokens re-explaining your project every session

Want their AI to remember debugging sessions from last week

Need context that follows them across different AI tools

Care about privacy — code never leaves their machine

Integrations

Works With Every AI CLI

One memory engine for all your AI tools. No configuration needed.

C
Claude

Anthropic

G
Gemini

Google

A
Aider

Paul Gauthier

Q
Qwen

Alibaba

l
llm

Simon Willison

s
sgpt

Community

m
mods

Charmbracelet

O
Open Interpreter

Community

C
Continue

Continue

C
Cursor

Cursor

C
Claude Desktop

Anthropic

FAQ

Frequently Asked Questions

No. All embeddings, AST parsing, and graph storage run locally in ~/.eidos/. There are zero cloud dependencies. Your code never leaves your computer.

Claude Code, Gemini CLI, Qwen, Aider, llm, sgpt, mods, Open Interpreter, Continue, Cursor, Claude Desktop, and any OpenAI-compatible tool. One memory engine for all of them.

EidosCore replaces claude-mem, memory, graphify, and caveman with one tool. It adds automatic context injection, token-budgeted assembly, session continuity (QMS), and self-learning retrieval — none of which claude-mem offers.

Yes. MIT licensed. Local-first. No cloud dependency. No hidden costs, no API keys needed for the core functionality.

In our debugging benchmark: 1,245 words down to 57 words — a 95.6% reduction. At 50 prompts/day, that's roughly $90/month per developer in token costs saved.

No. Context retrieval completes in under 10ms. You won't notice any delay in your AI coding workflow.

Roadmap

What's Next

We're actively building. Here's what's coming.

Universal CLI adapters (9+ tools)
Done
MCP server with 14 tools
Done
Web dashboard with knowledge graph
Done
Browser extension (ChatGPT, Claude.ai, Gemini web)
Done
JetBrains plugin
In Progress
Team analytics dashboard
Planned
VS Code extension
Planned

Give Your AI a Memory

One install. Every AI CLI. Zero config. Your AI remembers everything.

Star on GitHub