Giulio Camillo · the Theatre of Memory, 1530sThe machine it needed did not exist — until now.

Waku Seed · 2026 · Confidential

Own your memory

WAKU

One memory that follows you across every AI harness.

Backed by Character.VC · github.com/ShenSeanChen/waku-agent

[00]Team

Meet our team

Founder

Shen (Sean) Chen

MIT · Operations Research & AI

Ex-Google · 10M+ Business Messaging

Ex-Walmart · AI PM, Merchandising & Assortment

Founding Backend Engineer

Ian Zhao

Ex-Amazon & AWS · Global B2B AI Backend Systems

Founding Design Engineer

Emily Tsai

End-to-end product & frontend · Former Business Ops & HR

[01]Traction

How we started: Waku Agent

Waku Agent reached 1,000 GitHub stars in 25 days. It reached 100 stars in the first 18 hours.

1,834GitHub stars

369Forks

28Contributors

2M+Organic reach

ShenSeanChen/waku-agent on GitHub — live repository card — Live from GitHub github.com/ShenSeanChen/waku-agent

[02]Distribution

Building in public

Audience became our first users

Our You Can Build Anything series teaches the stack Waku is built on. 2M+ organic reach turns into installs, stars and contributors.

youtube.com/@SeanAIStories

Agent Harness & Loop Engineering 352K views AI Agent System Design 193K views AI Agent Memory System 66K views AI Agent Graph Engineering 58K views Hermes Agent Harness 41K views Waku Agent: harness in real code 35K views

[03]Proof

Proof

Don’t take our word for it

Nous Research (Hermes) and Pi replied to our videos. Supabase featured our open-source template.

Nous Research (Hermes): ‘We love this guy’s videos!’

“We love this guy’s videos!”Nous Research

Pi (@pidotdev) reacts to Sean’s Pi deep-dive

Reacted to our deep-divePi

“Made with Supabase”Supabase

[04]Reposts

Proof

The category reposts our explainers

Five AI accounts on X reposted our explainer videos. The five reposts drew 1.9M views in total. Waku Agent reached 1,834 GitHub stars in 77 days.

@0xMovez566K views

@precisox449K views

Praise for Sean’s explainers — 399K views

@cyrilXBT399K views

@LunarResearcher351K views

Our reply336K views

@0xCodez173K views

Hermes on LinkedIn41.3K views

Sean’s original open-source announcement

Our open-source launchOriginal

Swipe sideways to see all 8 posts.

[05]Problem

Problem

Your context outgrows your tools

01Migration friction

Explain again→ Copy-paste→ Give up

AI models don't share memory. Switching tools feels like onboarding a fresh intern every time.

AI tools don't share memory. Each new tool knows nothing about you.

02Decision hallucination

Sounds right→ Actually wrong→ Double-check everything

The AI keeps using facts that are no longer true. The AI confidently executes using outdated rules and bad context.

The AI keeps using facts that are no longer true.

You lose months of teaching when you switch to another tool. The AI makes later decisions from outdated notes.

[06]Flywheel

The revenue loop

Free local adoption drives memory accumulation.

Monetization triggers automatically when users hit multi-device, multi-app, or team collaboration limits.

01Local-first adoption Free local agent drives zero-cost memory capture.

02Memory accumulation More usage builds a richer personal memory graph.

03Sync & sharing limits Local-first fails when users need multi-device or team access.

04Paid cloud hosting Users pay for cloud sync, audit logs, and web UI.

Better recall on old memory

[07]Market

Beyond human

AI agents are the fastest-growing users of AI. Most companies buy their AI tools instead of building them.

Agent adoption

7.3T

Agent tokens in August, up from 0.51T in February

5times the growth of human use over the same period

OpenRouter data, Aug 2026

Enterprise deployment

88%

Enterprises with AI in regular use

76%buy AI rather than build it

McKinsey, Nov 2025 · Menlo Ventures, Dec 2025

[08]Why Waku · 1 of 3

Portability

One memory, every harness

Stored once. Handed over in the shape each tool reads.

[09]Why Waku · 2 of 3

Self-maintaining

Waku removes a fact when a newer decision replaces it

01You “We authenticate backend requests with JWT.”

02Waku writes it down as a fact, with who decided it and when Addauth standard · JWT

03You three months later, in another session: “Alex has moved us to OAuth2.”

04Waku sees the new decision contradicts the stored one, and stops handing the old rule over Retireauth standard · JWTAddauth standard · OAuth2

05You disagree with either one? edit or delete it — the entry is yours

Every fact carries who decided it and when — and a person can overrule it. The record is the team’s, not the model’s.

[10]Why Waku · 3 of 3

The differentiator

Waku writes down team decisions while people work

Nobody on the team has to stop and write a decision down. Waku writes the decision down instead.

[11]Landscape

Competitive landscape

Team knowledge Personal recall Per-harness Every harness

NotionConfluence Curated docs a human has to go and read

Claude CodeCursorChatGPT Memory locked to the tool that made it

Mem0ZepLetta Vector recall, personal, no review step

WAKU Reviewed team lessons reach every AI tool.

Team knowledge · per-harnessNotionConfluenceCurated docs a human has to go and read.

Personal recall · per-harnessClaude CodeCursorChatGPTMemory locked to the tool that made it.

Personal recall · every harnessMem0ZepLettaVector recall, personal, no review step.

Team knowledge · every harnessWAKUReviewed team lessons reach every AI tool.

[12]Where this goes

Roadmap

From one developer to the enterprise

One memory layer, three buyers. Adoption starts free and local, and every step up is a paid one.

Today — the individual

Memory installs into the agents a developer already uses. Free local, paid cloud.

Next — the team

Shared spaces and reviewed decisions: one memory behind every teammate’s agent. Priced per seat.

Then — the enterprise

Your models, your memory, your harness — inside your own boundary.

[13]The ask

Fundraising

$3M Seed

~70% team · 15% infrastructure & model costs · 10% go-to-market · 5% operations. Roughly 18–24 months of runway.

Goal for the next 18 months

Team

Hiring 2–3 engineers into a five-to-six person team building the memory layer.

Product

Automatic capture generally available across every major coding harness, shared team memory, and the enterprise controls that unlock paid seats.

[14]Contact

Contact

Looking forward to talking further

Founder · Waku

Shen (Sean) Chen

Email

sean@waku.one

Elsewhere

ShenSeanChen @ShenSeanChen @SeanAIStories shen-sean-chen