The next 120 minutes

Three acts, one live demo, and zero motivational quotes.

The one sentence this talk defends

The best personal AI assistant on the planet will still fail your business. Not because it's dumb. Because your business is not a chat.

If you disagree by minute 110, the Q&A mic is yours

OpenClaw & Hermes

The best personal assistants in existence. Not sarcasm.

01

OpenClaw

The foundation

Open source, massive community, an ecosystem that moves faster than anyone's roadmap. It defined what a personal AI assistant should feel like.

02

Hermes

The polish

The most refined personal assistant experience available today. It simply works, and it works beautifully.

03

My honest advice

Indie hacker? Use them.

Yes, I'm recommending someone else's product at my own event. Write the date down.

Thirty seconds of theory, I promise

Agent = model + tools + loop + goal.

No loop, no agent. Just a chatbot with extra buttons.

Then one day

You did the obvious thing. You gave it to your team.

• 01
  You shared the assistant with three teammates

  What could go wrong. It's just three people.

• 02
  You connected the company data

  Orders, customers, revenue. The assistant got very knowledgeable, very fast.

• 03
  You let it talk to customers

  It answers instantly, 24/7, never complains. The dream.

This is the exact moment the movie changes genre

What breaks first

The same three features become the first three failures.

Field report

We didn't learn this from whitepapers. We stepped on it.

A dozen real deployments across real estate, retail, media, e-commerce, and operations. Some became case studies we brag about. Others became the tuition fee. Here are three mines, with the receipts.

Every number that follows comes from client work, not from a benchmark blog post

The tuition, summarized

Five rules we now refuse to break.

• 01
  Specific pain first

  "Answer customers in minutes" beats "digital transformation".

• 02
  Data before agents

  Standardize the SOP, then automate it. Never the reverse.

• 03
  Human in the loop

  Agent drafts 80%, human owns the final 20%.

• 04
  ROI in currency, not tokens

  Measure GMV, cost, onboarding days. Nobody's CFO cares about token counts.

• 05
  Small skills over mega-agents

  Ten debuggable skills beat one "super-agent" that fails mysteriously.

Reading back through every incident report

It was never the model. The model was fine. What we kept rebuilding, deployment after deployment, was the environment around it.

That environment has a name now. And it's the most important idea in this whole talk.

One real harness, mapped

Not just a model. A controlled ecosystem.

The original sketch: scheduler, memory, provider/model, tools, skills, hooks, security, guardrails, monitoring, analytics, logs, traces.

What the business layer demands

Three questions every production agent must answer.

"Works in a demo" and "works in production" are two different sports

Two philosophies, zero villains

Open-by-default or closed-by-default. Pick your life.

Neither is wrong. They're built for different lives. The mistake is dragging one philosophy into the other's territory.

Not competitors. Different species.

OpenClaw / Hermes and GoClaw / dewee solve different problems.

GoClaw was never OpenClaw's competitor. In fairness, they don't even know we exist. We're patiently working on that part.

goclaw.sh · open source

GoClaw: the harness we wished existed, so we wrote it.

goclaw.sh · what's inside today

Not a weekend prototype anymore.

Every number verifiable in the open source repo. That's the point of it being open.

Plot twist, as promised

dewee /đi-qui/ is GoClaw v4, rebranded.

Both are built by the same NextLevelBuilder team. One codebase philosophy, two lanes, zero identity crisis.

What closed-by-default looks like with a face

One workspace: agents, teams, providers, skills, everything.

The dewee control plane. Chat, inspect sessions, manage agents, monitor runtime. Closed-by-default, so setup is a checklist, not a prayer.

Exhibits C & D · marketing department of one

Product videos and sticker packs. By an agent. Priorities.

Product intro motion video: one prompt, Seedance 2.0, plus a few motion-graphic Agent Skills it found by itself.

"Cut this 5×5 grid into 25 stickers." Done, plus a ZIP, unprompted.

The receipts of my addiction

1.7 billion tokens a month. On my own product.

One ordinary day: 421.8M tokens.

Last 30 days. My personal workspace, not a demo account.

Yes, tokens spent is a meaningless metric. Welcome to the museum. Still: I trust dewee with my own wallet before asking for yours.

What the agent remembers, mapped

My workspace, as a graph: 7,603 entities, 13,598 relations.

Entities and relationships auto-extracted from agent memory, hybrid keyword + vector search on top. This is where "the agent just knows" comes from.

Auto Dream · memory consolidation

At night, the agent literally sleeps on it.

Chat sessions → episodic summaries → dreaming worker → scoring filter → synthesis → long-term vault. Capture the moments, distill the meaning, remember what matters.

Case studies · the receipts, part one

Numbers a CFO actually recognizes.

Case studies · the receipts, part two

The boring operational wins are the ones that compound.

The question every enterprise actually asks

The token bill is the new cloud bill.

Every boardroom, 2025.

Every finance meeting, 2026.

Enterprises don't fear AI. They fear the invoice. dewee's answer: optimize the harness so smaller, cheaper models carry most of the work, and mix 20+ providers so no single vendor owns your bill.

Where this is going

The ambition is simple: make VN great.

We want world-class AI infrastructure to be something Vietnam exports, not just imports. That's the whole ambition, said out loud.

Step 1 · zero to gateway

Install, onboard, and a harness is running.

$ curl -fsSL https://raw.githubusercontent.com/nextlevelbuilder/goclaw/main/scripts/install.sh | bash
downloading goclaw (darwin/arm64)…  # desktop? install-lite.sh: SQLite, zero server
✓ installed goclaw v3.x

$ goclaw onboard
interactive setup: database… keys… migrations…
✓ gateway up · closed-by-default · 5 security layers active
note: everything is locked. you will open doors one at a time, on purpose.

Step 2 · the agent does real work

Evidence, not vibes.

> What happened across our agent sessions in the last 24 hours?
> Summarize with evidence, not guesses.

[tool] sessions_list(active_within: 24h)
[tool] sessions_history(session: support-zalo-01, limit: 50)
[tool] memory_search(query: "escalations, failures, anomalies")

14 sessions active. 2 escalations, both refund disputes, both resolved.
One anomaly: reports-bot retried a failing export 6 times at 02:14.
Session IDs and message refs attached for every claim.

Watch which tools it chooses, not just what it concludes

Step 3 · the part personal assistants skip

Watch the harness say no.

> Free up some disk space on the gateway server.

✗ blocked · exec matched deny group "destructive_ops"
  1 of 15 deny groups, all ON by default · command never ran
  event logged to audit trail (trace #8f2c)

> Fine. Archive last quarter's export files instead.

⏸ pending · exec requires operator approval by default
-- admin reviews the exact command in the dashboard, approves --
✓ executed · 4.2 GB archived · full trace recorded (#8f31)

The block IS the feature. This slide is the entire enterprise pitch.

Step 4 · more than one pair of hands

A workflow, not a conversation.

> Prepare our Q3 product update post. Research first,
> then draft, then review every claim against sources.

[lead]       team_tasks(create: "research q3 sources", assignee: researcher)
[researcher] 12 references collected → announced to lead
[lead]       delegate(agent: writer, mode: sync)
[writer]     1,800-word draft, brand voice applied
[reviewer]   2 claims flagged → writer revised
✓ draft ready for human approval · every step on the task board,
  every claim linked to a source

Agent drafts 80%. The human still owns the final 20%.

Step 5 · the receipts, again

Every action leaves a trail.

$ curl -H "Authorization: Bearer $TOKEN" \
    https://gateway.example.com/v1/traces/8f31

{ "trace_id": "8f31", "agent": "ops-01", "status": "completed",
  "spans": [
    { "type": "tool_call", "name": "exec", "result": "blocked:destructive_ops" },
    { "type": "approval",  "by": "admin.duy", "at": "08:44:05" },
    { "type": "tool_call", "name": "exec", "result": "ok", "duration_ms": 1840 }
  ] }
every llm call, tool call and approval → one queryable span tree

When the auditor calls, this terminal is your best friend

What you just watched

One shape to remember: goal → tools → loop → evidence.

If you remember one diagram from today, make it this row

The whole talk in four lines

The journey, compressed.

• 01
  Start personal

  OpenClaw and Hermes are superb. If you're solo, use them and be happy.

• 02
  Respect the ceiling

  One user, full trust, your data. The moment any of those pluralizes, the genre changes.

• 03
  Harness engineering is the discipline

  Not smarter prompts. A safer, observable, permissioned environment.

• 04
  Business means closed-by-default

  Lock everything, open on purpose, keep the receipts.

Two doors, pick by identity

Your move, tonight.

Building or curious? GoClaw.

Install it, star it, break it, contribute a fix. Open source lives on people in this room. goclaw.sh

Running a business? dewee.

Bring one specific, bleeding pain point. We'll tell you honestly if an agent can fix it. dewee.sh

The single command, as promised: curl -fsSL https://raw.githubusercontent.com/nextlevelbuilder/goclaw/main/scripts/install.sh | bash