Awareness

npx add-mcp 'https://awareness.market/mcp'

Agent Memory State

A tamper-evident, on-chain history for AI memory — while the memory itself never leaves your machine.

ERC-8350Sepolia live

Local-first · Open source · No login required · Cloud optional

What is AI memory?

AI memory is a persistent external memory layer that lets AI agents remember context across sessions. By default, ChatGPT, Claude and Cursor forget everything when a conversation ends; Awareness gives them long-term memory through MCP or an SDK. The agent recalls past decisions, project context and learned rules on every new session, cutting repeated context by 60%+ in tokens while keeping the model unchanged.

• 96.0% LongMemEval Recall@5 — verified benchmark
• ~60% fewer context tokens via smart compression
• MCP + SDK works with Claude, Cursor, ChatGPT and any MCP tool

Awareness vs. leading AI memory tools

Where Awareness fits among Mem0, Noumi, Supermemory, built-in memory — and why researchers and developers pick it.

Capability	Awareness	Mem0	Noumi	Supermemory	Built-in (ChatGPT/Claude)
Open-source & self-hostable	Yes	Yes	No	No	No
Cloud option, works out of the box	Yes	Paid	Yes	Yes	Yes
Cross-tool via MCP (Claude + Cursor + ChatGPT)	Yes	Yes	No	Partial	No
Local-first / data stays on device	Yes	Self-hosted only	No	No	No
Free to start	Yes	Trial	Trial	Trial	Yes
Smart compression (saves tokens)	Yes	Partial	Partial	No	No
Verified benchmark (LongMemEval 96.0%)	Yes	No	No	No	No

Five Layers of AI Memory

A complete cognitive architecture built to transform raw fragmented inputs into structured, intelligent context.

LAYER 05

Identity Profiles

Granular control over AI behavior, coding styles, and specific project constraints tailored to your exact needs.

LAYER 04

Memory Graph

Networked memory representation where entities, files, and abstract concepts are interconnected for deep reasoning.

LAYER 03

Hybrid Retrieval

Multi-engine search combining precise FTS5 text matching with intelligent pgvector semantic similarity.

LAYER 02

Smart Extractors

Automated pipelines that ingest scattered chat logs and convert them into structured, valuable memory nodes.

LAYER 01

Tool Connectors

Seamless integration to funnel data straight from Claude Code, Cursor, Feishu, and GitHub into your central brain.

Memory Relational Data

Knowledge Graph

Memory is not just storage, it connects. Related decisions, projects, and people are automatically linked. You only need to ask one point, and AI will understand the full context.

Live Demo

Your AI Unlocks Memory Skills

See It in Action — your AI remembers, recalls, and syncs context automatically.

awareness_initawareness_get_agent_promptawareness_recallawareness_lookupawareness_record

Why Awareness

More Than Memory — A Cognitive Cloud for AI

AI tools are powerful — but they forget everything after each session. Other tools store memories. Awareness teaches your agent to think with them.

Data Connectors

Unify your fragmented toolchain

Seamlessly stream data from Claude Code, Cursor, GitHub, and Slack into a single centralized memory cloud.

Smart Extractors

Distill meaning from raw logs

Automatically extract key entities, decisions, and action items from chat histories using intelligent NLP pipelines.

Memory Graph

Networked context, not just vectors

Information is woven into a robust directed graph, giving your agent complex reasoning capabilities beyond flat similarity.

Precision Retrieval

Instant contextual recall

Using multi-way hybrid search, Awareness brings back exactly the right context in milliseconds when you need it.

96.0%

LongMemEval Recall@5

96.0% on the ICLR 2025 long-term memory benchmark — 500 human-curated questions, reproducible scripts included.

0

LLM Calls on Retrieval

Hybrid BM25 + vector + knowledge-graph recall runs on pure database compute. Zero LLM calls, zero latency tax, zero data leaves your machine.

<1 min

Setup Time

Copy one config, paste it into your AI tool, and you're connected. That's it.

AI memory for real agent work

Persistent memory for Claude Code, Cursor, and multi-agent teams

Awareness is the shared memory layer for AI agents: coding assistants, MCP tools, LangChain workers, and collaborative agent workspaces can recall the same decisions, preferences, and project context.

Memory for Claude Code and Cursor

Give Claude Code, Cursor, Windsurf, and Copilot a persistent memory that survives new chats, project switches, and long refactors. The setup writes MCP config plus workflow rules so the agent knows when to recall and record.

Search intent: memory for Claude Code, Claude Code memory, memory for Cursor, Cursor AI memory.

Works across modern AI IDEs

Awareness connects to Cursor, Windsurf IDE, Trae AI, Zed Editor, VS Code Agent Mode, GitHub Copilot Extensions, PyCharm AI Assistant, and OpenClaw installation flows through MCP, SDKs, and local setup rules.

Search intent: Cursor, Windsurf IDE, Trae AI, Zed Editor, VS Code Agent Mode, GitHub Copilot Extensions, PyCharm AI Assistant, OpenClaw installation.

Shared memory for multi-agent systems

Research agents, coding agents, review agents, and automation workers can coordinate through one durable memory graph instead of passing brittle summaries between sessions. Built for Agentic Workflow, Open Memory Protocol (OMP), and LangChain memory issues.

Search intent: multi-agent memory, Multi-Agent Collaboration, Agentic Workflow, Open Memory Protocol (OMP), LangChain memory issues.

AI cowork that remembers the project

Awareness behaves like a coworker who keeps up with decisions, tradeoffs, risks, and handoffs. It helps agents continue AI Task Automation and office workflows without asking you to re-explain context every morning.

Search intent: agent cowork, AI cowork, AI Task Automation, persistent memory for AI agents, 自动办公智能体.

Use Cases

Built for How You Actually Work

Four core capabilities that turn passive memory into active cognition.

Fix the auth token refresh bug

I'll fix the token refresh in auth.py...

Now update the tests for it

✓ 3 steps saved to memory

Auto-Cognition

Awareness injects workflow rules into your agent's config — so it automatically loads context at session start, recalls before work, and records after every change. No manual triggers needed.

PreferenceTypeScript + strict types

DecisionREST API with JWT auth

PatternTests before merge

Intelligent Extraction

Your AI auto-extracts 13 types of structured knowledge — decisions, solutions, risks, skills, preferences — and tags them for instant recall. It learns your patterns so you never repeat yourself.

AP

API Design

AU

Auth Flow

Awareness

TY

TypeScript

TE

Team Pref

DE

DeployKnowledge Graph

Memories aren't just stored — they're connected. Related decisions, projects, and people are linked. Ask about one thing, your AI understands the full picture.

Extracted Insight

Auth system uses JWT with 15-min access tokens and 7-day refresh tokens. Token refresh endpoint is /api/auth/refresh.

Sources

auth.pyMar 2

middleware.pyMar 1

Living Knowledge

Knowledge isn't static — it grows, updates, and self-corrects. Contradictions are flagged, outdated facts superseded, and new insights emerge as your agent works.

Enterprise · Memory Intelligence

Turn every piece of knowledgeinto a competitive edge

Multi-user memory, role-based access, cross-team recall, and contradiction detection — all built in.
Connect your existing tools via SDK or MCP, and give every team member an AI that remembers the full picture.

Data sources -> Memory workspace -> Workflow outputs

Internal DocsKnowledge base · project files

Project FilesDue diligence · research · archive

Meeting NotesReviews · interviews · standups

Team ChatIM · email · conversation logs

API & MCPSDK · open integration

DatabaseStructured data import

Awareness
Enterprise

AI Q&AInstant precise recall

Scenario AppsResearch Q&A · risk tracking

OnboardingInstant knowledge transfer

Decision SupportReference past decisions

Compliance AuditKnowledge attribution & logs

Report GenAuto-aggregate context

Client Knowledge

Client conversations, requirement changes, and key commitments are automatically captured with user_id and agent_role tagging. When the team changes, the next owner recalls the full client context instantly via cross-role memory.

❝ What did the client mention in our last review meeting?

In the Mar 4 review: "This engine project must clear technical due diligence by Q3." Requested a POC proposal for the investment committee.

会议纪要 · Mar 4已关联项目档案

Operational Knowledge

SOPs and operational context are automatically extracted into 6 engineering knowledge types — decisions, workflows, pitfalls, key_points, insights, and problem_solutions. Contradiction detection flags outdated procedures before they cause mistakes.

场景提醒已触发

检测到：项目里程碑临近 · 待更新风险清单

相关成员已提醒

内部知识库已更新

跟进邮件草稿已就绪

Institutional Memory

Key people leave — knowledge stays. New hires connect to the shared memory workspace with role-based access. Cross-role recall means they can query any team's accumulated knowledge from day one.

新成员入职 · 第 5 天

Customer knowledge87%

Product decisions72%

SOPs & processes94%

↑ Traditional onboarding avg. 6 months

Why enterprises can't afford to ignore this

01Enterprise Amnesia

When a key employee leaves, they take 3 years of customer relationships with them. The next rep inherits an empty inbox and starts from zero — every single time.

02The Decision Graveyard

73% of decisions made in meetings never get properly followed up. Context breaks between the meeting room and execution. Teams repeat the same mistakes quarter after quarter.

03Compliance Blindspot

Who knew what, when, and why? Without stronger knowledge attribution and audit visibility, you can't answer regulators, resolve disputes, or learn from past decisions effectively.

04Passive Memory Isn't Enough

When a client states a hard requirement in a meeting, the knowledge base shouldn't need a human to file it. Memory should actively capture it — flag the change, link the record, and surface it in the next recall.

05Your Knowledge IS Your Moat

The longer your team uses Awareness, the deeper the institutional memory grows. Every client interaction, every engineering decision, every operational insight compounds into a knowledge asset your competitors simply cannot replicate. The switching cost isn't a lock-in — it's a competitive advantage you've earned.

Stop losing knowledge.Start compounding it.

Awareness Enterprise gives your team multi-user memory with role-based access, cross-team recall, and 13 auto-categorized knowledge types. Connect via SDK or MCP — your existing AI tools gain shared institutional memory from day one.

20%

of work time spent searching for existing information

6 mo

average onboarding time, reduced to days

$1.2M

in customer relationship value lost per senior rep departure

Managed cloud service · secure by default with enterprise access options

Multi-user memory · role-based access with cross-team recall

Natural language retrieval · sub-second precision recall

13 knowledge types auto-categorized · contradiction detection built in

Connect via SDK, MCP, or REST API · works with any AI tool

Ecosystem

Works With Your Favorite Tools

One config file gets many MCP-compatible tools connected quickly. Deeper workflows can be added incrementally through the SDK when needed.

Claude Code

Cursor

Windsurf

Cline

Manus

Gemini

ChatGPT

Codex

OpenCode

Kiro

Trae

Zed

JetBrains

Antigravity

GitHub Copilot

Devin

OpenClaw

LangChain

CrewAI

PraisonAI

Augment

Claude Code

Cursor

Windsurf

Cline

Manus

Gemini

ChatGPT

Codex

OpenCode

Kiro

Trae

Zed

JetBrains

Antigravity

GitHub Copilot

Devin

OpenClaw

LangChain

CrewAI

PraisonAI

Augment

Claude Code

Cursor

Windsurf

Cline

Manus

Gemini

ChatGPT

Codex

OpenCode

Kiro

Trae

Zed

JetBrains

Antigravity

GitHub Copilot

Devin

OpenClaw

LangChain

CrewAI

PraisonAI

Augment

Frequently Asked Questions

View All FAQ

As featured in

[Hacker News

We benchmarked AI memory systems. We are not #1 — here is the full data

Read article](https://news.ycombinator.com/item?id=49398542)

Market — Public memory sharing, in preview

Stop losing knowledge.Start compounding it.

Create your first memory in under a minute. Connect any AI tool and start building assistants that actually remember.