@i5ting: Three approaches to implementing digital employees: 1) Alibaba/Tencent-style: running in Docker with full permissions, limited by Docker, can only solve 20% of needs at best. I don't like this. 2) IM-based: putting the agent in contacts like bloome, running on PC, sharing environment and context with the host, etc.
Summary
In a tweet thread, three approaches to implementing digital employees are discussed: running in Docker, IM-based agents (e.g., bloome), and digital twin-based solutions (e.g., WisMe.ai). It also introduces WisMe.ai as a personal AI knowledge base and reading assistant product.
View Cached Full Text
Cached at: 07/11/26, 03:27 PM
There are three approaches to implementing digital employees:
- The Alibaba/Tencent approach: running in Docker with various permissions enabled. Its limitation lies in Docker; it can solve at most 20% of needs. I don’t like this.
- IM-based approach: putting the agent as a contact, like Bloome. This runs on a PC, sharing the same environment, context, etc., with the host. Interaction is very convenient. This kind is very helpful for daily use, covering more than half of needs.
- Human digital twin approach: it senses everything you see and do, distills it, similar to http://wisme.ai. This is about distilling yourself. When it’s used as an agent appearing in an IM, it becomes the most realistic and powerful.
WisMe.ai — Your personal AI knowledge base & reading assistant
Source: https://wisme.ai/en WisMe.ai (https://wisme.ai/en)PERSONAL KNOWLEDGE OS · β
VOL. 01 · ISSUE 04 · SPRING 2026§ PROLOGUE
CHROME EXTENSION · PRIVATE BETA
What you read, becomesyou.
A quiet companion that reads what you read, and weaves it into a knowledge of your own.
WisMe.ai is an AI research companion that lives in your browser. Itquietly recordswhat you read each day, distills what is worth remembering, builds your personal knowledge graph, and keepscollecting, researching, and connectingwhile you sleep.
12,847
pages recorded
across private beta
3.2M
knowledge nodes
extracted & linked
97%
recall accuracy
on user-marked items
your.graph / live
● RECDAY 1/7
AI Systems
+ NEW NODE
“Zettelkasten” captured from arxiv.org
∞ LINK DISCOVERED
RAG ↔ Long-term Memory
§ 01HOW IT WORKSFOUR QUIET ACTS
Not another note app. AnAI that reads with you.
“Knowledge should accumulate like compound interest — without compound effort.”
— DESIGN PRINCIPLE
STEP 01Observe
REC
Observe
The extension runs quietly in the background and records the pages you actually dwell on, revisit, and scroll. Noise is filtered out.
A quiet observer. Only pages you actually read — measured by dwell, scroll, re-visits.
STEP 02Distill
ENTRY · 04.17
Distill
Every night the AI reads your day, extracts key concepts, claims, data, citations, and writes a structured “daily knowledge entry”.
Each night, AI re-reads your day and extracts concepts, claims, data, citations.
STEP 03Weave
Weave
New entries are woven into your graph. The AI finds cross-day and cross-domain links, surfacing neighbouring ideas you might have missed.
Entries weave into your graph. Cross-day, cross-domain connections surface as you sleep.
STEP 04Research
?
Research
Give it a goal and the AI searches the web, traces citations, compares viewpoints, and writes the research report that’s still missing.
Give it a question. It searches the web, chases citations, writes the missing piece.
§ 02FEATURESEIGHT AFFORDANCES
Don’t change how you read — just make readingthicker.
01/ 08Silent Capture
Silent Capture
The extension recognises pages you actually read and ignores the fleeting noise.
02/ 08Interest Radar
Interest Radar
The AI infers what you’ve been focused on recently and flags important items you keep skipping.
03/ 08Daily Digest
Daily Digest
Delivered before 6 am the next morning: a distilled “what you read” with key points, new concepts, loose threads.
04/ 08Living Graph
Living Graph
Concepts, papers, people, events auto-connect. Trace by time, domain, or source.
05/ 08Serendipity
Serendipity
Rediscover a passage you read three months ago that happens to answer today’s question.
06/ 08Research Agent
Research Agent
Hand it a topic; the AI searches, compares, summarises, and returns a cited report.
07/ 08Web Reach
Web Reach
When your library lacks a piece, the AI crawls arxiv, Wikipedia, and domain sites to fill it.
08/ 08Secure by Default
Secure by Default
Data is encrypted in transit and at rest; common sensitive fields are redacted before storage.
§ 03THE LIVING GRAPHHOVER TO EXPLORE
AI SystemsRAGAgentsLong-term MemoryPKMZettelkastenKnowledge GraphEmbeddingsOntologySerendipityCitation NetSummarization
FIG. 03 — A ONE-WEEK SLICE OF YOUR GRAPH↓ DRAG · CLICK · FILTER
Every node carries aprovenance.
Click a node and you see: when and from which article and passage it was extracted; which neighbours it links to; and a one-line “why it matters” generated by the AI.
Nodes
concepts · people · papers · orgs
3,247
Edges
typed: cites, contradicts, extends, …
9,812
Sources
arxiv · ssrn · blogs · twitter · books
412 sites
Clusters
auto-detected by embedding density
27 themes
§ 04DAILY DIGESTHOVER ITEMS FOR DETAIL
A morning brief of what youread yesterday.
Like a personal Economist — but every item comes from your own reading yesterday. It keeps what is worth remembering, nudges you on what’s unfinished, and surfaces connections you may have missed.
DELIVERED BY 6 AM DAILY VIA EMAIL · BROWSER · API ~3 MIN READ · 8–12 ITEMS
THE DAILY · VOL. 214
Thursday, April 17
14 PAGES READ
est. 2h 18m · 6 domains
HEADLINE INSIGHT
“The three papers you read today share an unspoken premise:memory is reconstruction, not storage. This echoes the Tulving 1972 piece you read in March.”
GENERATED FROM · arxiv:2504.11203 · 2504.10998 · sciam.com
NEW CONCEPTS · 4
01
Reconstructive MemoryHIGH
Memory is active reconstruction, not passive replay
arxiv:2504.11203
02
Semantic DriftMEDIUM
Meaning of a concept drifts across a corpus
acl anth.
03
Epistemic HumilityMEDIUM
A deliberately trained cognitive humility
lesswrong
04
Hippocampal IndexingHIGH
Hippocampus as an index to memory — a hypothesis
nature
UNFINISHED · 3
The Scaling Hypothesis, Revisited
Making It Stick, ch. 7 — Meta-organization of knowledge
Memory, Attention and Prediction
§ 05RESEARCH AGENTLIVE DEMO
Hand it a question; it comes backunderstanding.
The agent first reads your existing library, then decides what else it needs externally. It chases citations, compares viewpoints, builds timelines, and delivers a cited, verifiable report you can keep questioning.
ASK THE RESEARCH AGENT
DEPTH: DEEP · SOURCES: 8–12 · ETA: ~3 MIN
RESEARCH.TRACE · LIVE
○ IDLE
PLAN
Decompose into 4 sub-questions and plan the search path
RECALL
Search your existing library
EXTERNAL
Launch external search & crawling
SYNTHESIZE
Compare viewpoints; identify consensus and divergence
WRITE
Draft a cited report
SECURE BY DEFAULT
Your reading, should belong only toyou.
All data is encrypted in transit and at rest; common sensitive fields are redacted before storage. You can permanently delete your data at any time.
READ THE PRIVACY CHARTER → (https://wisme.ai/en/privacy)
OUR COMMITMENTS
01End-to-end encryption in transit & at rest
02Automatic redaction of sensitive content
03Permanent delete on request
04SOC 2 & GDPR aligned
§ 06PRICINGSIMPLE · TRANSPARENT
Pick the pacethat fits you.
Free includes 50 credits and a daily Lite report, enough to start capturing and running lightweight page conversations. Pro includes 1,000 monthly credits for AI chat, Workbench nodes, and Skills; daily reports are included in the subscription.
Start accumulating
Start accumulating.
- →Unlimited browser extension
- →50 credits every month
- →Daily Lite report
- →Usually enough for 100+ lightweight page conversations
Let reading compound
Let reading compound.
LIMITED OFFER19\.9014.50USD
/ month
- →Everything in Free
- →1,000 credits every month
- →AI chat, Workbench nodes, and Skills consume credits
- →Daily report cost is included, no extra credits charged
- →Knowledge graph and long-term history retention
- →Cancel anytime; access remains until period end
Autonomous research, always on
Autonomous research, always on.
$100.00USD
/ month · tentative
- →Everything in Pro
- →Higher monthly credit allowance
- →Multimodal generation allowance
- →Automated research workflows
- →Priority queue
- →Team and heavy research support
USD / CNY PRICING · BILLED MONTHLY · CANCEL ANYTIMEEDUCATIONAL / NONPROFIT DISCOUNTS AVAILABLE →
§ CODA
Start letting your reading actuallyaccumulate.
Install the extension. Close this tab. Read as you always do. See what you’ve become in one week.
NO CREDIT CARD · CHROMIUM · EDGE · ARC · BRAVE
yan5xu (@yan5xu): Yesterday, I discussed AI employees with a friend from Microsoft, and we talked about digital twins. In my view, a digital twin can be part of an AI employee, but they are not fully equivalent. Only when something is enough to change existing production methods and organizational structures can it be called a revolution. Building digital twins for each position based on the current organizational structure is clearly limited by the present.
Similar Articles
@CoderDaMing: The Most Insane Startup Trend of 2026: Solo Founder, 0 Employees, $300K in Revenue! What Starting a Business Used to Require: Hire People → Rent Office → Team Bonding → Burn Cash for 6 Months → Still Not Live. Now All You Need Is: One Person + One http://Bloome.im Group with 5 Agen…
The author shares their experience of building a team of 5 AI Agents (Product Manager, Engineer, Designer, Copywriter/Operator, Data Analyst) on the Bloome.im platform to achieve $300K in annual revenue as a solo founder, highlighting that Agents can collaborate and delegate tasks to each other, with 20u of free tokens provided daily.
@blueskylh1: The most painful thing about solo product development or leading an AI team is being a "mindless messenger" between different chat windows. After the PM writes the requirements, I have to copy and paste them into the developer's chat. After seeing the sharing from Jason @jxnlco, a developer experience engineer on the OpenAI Codex team, I set up a workflow without...
Introduces a multi-AI agent collaborative workflow based on local plain text files and OpenAI Codex, allowing PM, backend, frontend, and QA to efficiently develop via file relay without copy-pasting.
@ai_laotie: https://x.com/ai_laotie/status/2068215413050347654
Introduces three AI workflows that will make money in 2026: reverse-engineering overseas case studies, converting long-form articles into viral X threads, and building a faceless AI short video matrix. Emphasizes that systematic workflows and prompt iteration matter more than chasing new tools.
@9hills: AI for Work is quite subtle inside many enterprises. Using general-purpose agents (Hermes, WorkBuddy, etc.) paired with the DeepSeek-V4-Pro model can optimize many repetitive tasks. But from what I observe, unlike programmers who actively burn tokens to improve efficiency and accelerate their own obsolescence…
A user observes the current state of internal AI applications: general-purpose agents with DeepSeek V4 Pro can optimize repetitive tasks, but traditional role departments often propose complex requirements that are hard to implement, while secretly using them privately, demonstrating a kind of survival wisdom.
@justloveabit: With This Open-Source Tool, I Got a Team of AIs to Work for Me. Here's the deal: I've been tinkering with various AI agents lately. Multiple Claude Code windows open, Codex running, occasionally using Cursor. The result? Total chaos—I had no idea what each agent was doing or how much it was costing. Restar…
This article introduces Paperclip, an open-source tool designed to centrally manage and orchestrate multiple AI agents. By simulating a corporate organizational structure, task assignment, and budget control, it addresses key pain points in multi-agent collaboration, such as lost context, unpredictable costs, and chaotic scheduling.