LP
01Lundy Pang
Co-founder & CEO · full-time
Prior founder, exited. Owns product, go-to-market, and the raise. Unbrowse is the only thing being built.
Unbrowse·Cover
01/ 13
SEED ROUND · CONFIDENTIAL
Unbrowse
The machine routing layer for the web
Agents should not browse websites.
They should route through them.
Scroll
02 · Why Now
Thewebisbecomingmachine-native.
For 30 years, websites were built for humans. Now agents are becoming the primary new users of the internet.
First
Browser agents proved it
Agents can act online. Anthropic Computer Use, OpenAI Operator.
Then
Web access went default
Every agent stack now ships with web access built in.
Now
A new bottleneck appeared
Browser execution is slow, expensive, and brittle. The web needs a machine-native layer.
The next internet layer is not better browser automation. It is machine routing.
$1T+ in annual machine web traffic is currently unrouted.
03 · The Problem
Browserexecutionisthewrongabstraction.
A browser is a machine pretending to be a human to talk to another machine. That works for demos. It breaks for repeated workflows.
10–45s
per action
Too slow for repeated workflows
$0.10+
per action
Too expensive at agent scale
Brittle
by design
UI changes break flows
Repetitive
by nature
Agents keep relearning the same paths
Agents do not need pixels, buttons, and DOM trees. They need routes.
04 · The Product
Unbrowseturnswebsitesintoreusablemachineroutes.
The first session discovers the route. Every later session reuses it.
Browser execution today
✗Open page
✗Parse DOM
✗Click through UI
✗Recover from failures
✗Repeat the same work again
Unbrowse execution
✔Discover route once
✔Call route directly
✔Return structured JSON
✔Reuse across future sessions
What Unbrowse returns
GET /2/timeline/home
{
"tweets": [{
"author": "sama"
}]
}
<1s · structured JSON
What takes browser agents seconds becomes a routed call in milliseconds. This is a different internet primitive.
05 · In Practice
Whatthislookslike.
An agent needs to look up a LinkedIn profile. Here is the difference.
Browser agent~10s · $0.30
1. Launch headless Chrome
2. Navigate to linkedin.com/in/target
3. Wait for page load (6s)
4. Handle login redirect (12s)
5. Parse 8,200 DOM nodes
6. Find name, title, company in HTML
7. Handle "verify you're human" popup
8. Extract text, hope layout didn't change
Tomorrow: LinkedIn ships a new layout.
Start over.
Unbrowse routed<1s · $0.005
REQUEST
unbrowse.resolve("linkedin.com/in/target")
RESPONSE
{
"name": "Jane Smith",
"title": "VP Engineering",
"company": "Acme Corp",
"location": "San Francisco"
}
Route discovered once. Reused by every agent after.
Same result. Up to 30x faster. 60x cheaper. No browser.
06 · The Wedge
Westartwherebrowserpainishighest.
Repeated multi-step agent workflows. Developers already run agents on the same sites every day. They feel the pain first and most clearly.
Research
Structured retrieval
Logged-in automation
Monitoring
Back-office tasks
Browser workflow: $0.30. Unbrowse routed: $0.005. The first wedge is obvious: replace repeated browser execution with reusable routes.
08 · Moat
Theroutinggraphisthemoat.
Each successful workflow leaves behind more than an execution.
Each workflow leaves behind
▸A route
▸A reusable machine path
▸A signal of what agents tried to do
▸Evidence of what worked
Machines need what humans got
▸Routes instead of browsers
▸Reusable execution paths instead of links
▸Reliable structured retrieval instead of search
▸Transaction rails instead of app stores
▸Site-level compensation
Every routed request trains the next one. Network request telemetry is a proxy for machine intent — what agents try to do, what works, what fails. This data compounds into a recommendation engine that predicts the next best route before the agent asks.
Google indexed the web for humans. Unbrowse routes it for machines.
09 · Business Model
Paidrouting,thenplatformeconomics.
Free today to maximize adoption. Paid once route reuse becomes embedded behavior.
$0.30
browser task (LLM + runtime + retries)
$0.005
Unbrowse routed task
60x cheaper
90–96% cost reduction per task
Browser task: $0.30 → Unbrowse routed: $0.005 → Platform take: $0.001
Example power team
Agents running5
Workflows per agent / day500
Routed requests per workflow8
Total routed requests / day20,000
Browser cost / day$6,000
Unbrowse cost / day$100
Customer saves98%
Platform revenue / month$600
1,000 teams = $7.2M ARR. 10,000 teams = $72M ARR.
10 · Market
Anewmarketisformingunderneaththeagenteconomy.
As machine traffic becomes a larger share of the web, the platform expands into:
Routing
What should the agent call?
Retrieval
How should it get the right data?
Execution
How should it act reliably?
Transactions
How should machine traffic pay?
Compensation
How do websites monetize agent usage?
Wedge
Repeated agent workflows
Highest pain, clearest ROI.
Expansion
Framework integrations
Every integration multiplies volume.
End state
Machine routing & transaction layer
Where agents go to use the internet.
This market grows with every new agent framework, every deployed workflow, and every website machines learn to use.
11 · Competition
Everyoneelsehelpsagentsclick.Unbrowsehelpsagentsroute.
Browser agents
Operator, Computer Use, Browser Use
Useful, but slow and brittle for repeated work.
Scraping platforms
Bright Data, Apify
Read-heavy, not built for reusable machine execution.
MCP / per-site tools
Hand-built integrations
Fragmented and impossible to scale across the open web.
Unbrowse
Machine routing infrastructure
Discovers routes, reuses them, compounds with usage. Open-sourced the client to commoditize competitors' moats. The network and routing graph cannot be replicated.
Others are tools. Unbrowse is infrastructure.
12 · Team
Builttowardthisbeforethemarketwasready.
LT
02Lewis Tham
Co-founder & CPO · creator
NUS Computing. Created initial iterations of Unbrowse in 2024. Authored “Internal APIs Are All You Need” (arXiv:2604.00694). Interviewed on This Week in Startups.
CC
03Cayden Chik
Co-founder & CTO · AI architecture
Lead AI architect, Singapore Army AI rollout. Research at NUS Deep Learning Lab. Previously MIT, Research Science Institute. Top 8 worldwide, International Mathematical Modeling Competition.
Why us
Eight iterations, one conclusion: browsers are the wrong abstraction for machines.
We are the users
We lived the discovery tax. The paper came out of the engine.
Backing
NUS Fintech Lab
Three founders, clean cap table, Singapore Pte Ltd.
Raising our seed
To build the default routing layer for agent workflows
Open on check size and structure. Let’s talk.
40%
Infrastructure
Route index, engine, graph expansion
30%
Distribution
Framework partnerships, developer adoption
30%
Team
Key hires. Lean by design
This round proves
✔Agents reusing routes across sessions, not just discovering them
✔Route reuse density compounding week over week
✔Paid workflows with design partners (HOYSAN live)
✔Open-source client driving organic adoption without sales
✔Bottoms-up path to venture-scale revenue
The browser was built for humans. Unbrowse is built for machines.
Seed Round · Confidential