Runs 24x7 on your machine

Turn your manual workflows
into AI agents

by Backed by Y Combinator

AI agents that can connect to 100+ integrations and automate manual tasks you couldn't automate before.

Runs local. No signup, no cloud, no per-seat pricing. Source-available, free to self-host.

Hiring for a whole team? Nex Cloud is the hosted big sibling. Get Nex Cloud →

How it works

Show. Build. Approve.

You are onboarding a hire, not configuring software.

01
Show
Describe the job in one sentence — or demo it once on a screen-share call while WUPHF watches.
02
Build
The agent assembles live in front of you: its own app, schedule, and tools.
03
Approve
Reads are free. Writes are held until you tap approve. Then it runs 24x7.
The demo call

The last time you do this job,
someone is finally taking notes.

A voice call with screen share. You do the job once. WUPHF asks the sharp questions, then builds the agent.

screen share · rec · Nex is watching and listening
YOU
opens the demo-request form, looks the company up in HubSpot, posts the good ones to Slack
NEX
Got it — you looked the company up in HubSpot and checked headcount.
NEX
Over what score does a lead go to an AE instead of the nurture list?
YOU
70 and up goes to #ae-handoffs with the reason. Rest gets nurtured.
NEX
I think I have the whole flow. Ready for me to build the agent from this, or is there more?
YOU
That's it. Build it.
NEX
Building Lead Routing Agent — app, routines, and the two tools it needs. Watch it come together.
terminal
# Option 1: run it
npx wuphf@latest
 
# Option 2: build from source
git clone https://github.com/nex-crm/wuphf.git
go build -o wuphf ./cmd/wuphf
 
# Browser opens at localhost:7891.
# Describe a job — or demo it on a call.
# Watch the agent build its own app.
Monday 9:00 · Routine: weekly pipeline recap
AGENTRan the recap. 3 stage moves, 2 new leads, Umbrella stalled 21 days.
AGENTDrafted the Slack post for #revops. Holding it for your approval.
YOU*one tap* Approved.
You demoed this job exactly once. You just trained your replacement — it says thank you and works weekends.
What you hired

Anatomy of your new hire.

Every agent ships with all six. Not a chatbot in a trench coat.

The app
A real screen
Its own UI, built live in front of you. Reads and writes real workspace data.
Routines
Scheduled work
"Every Monday 9:00." Versioned prompts, run history, a transcript per run.
Tools
Self-authored
"Score a lead." "Post to #ae-handoffs." It writes its own tools from what you taught.
Knowledge
Cited pages
Every claim carries a citation. Your old wiki and notebooks come along verbatim.
Data + integrations
Your whole stack
Its own typed tables, plus 100+ integrations via Composio. Connected once.
Approval gate
Sends with permission
No email, Slack post, or CRM write leaves without a human tap.
What people hire first

Agents your pipeline is already asking for.

Closed-lost re-engagement
Monitors deals lost 90+ days ago and detects what changed.
Saves ~$300–400/mo
CRM hygiene
Finds duplicates, missing fields, and stale records.
Saves ~$200–500/mo
Meeting prep
A cross-system account brief before every meeting.
Saves ~$100–200/mo
Deals going dark
Detects silence across email, Slack, and calendar on active deals.
Saves ~$200–400/mo
Lead scoring
Scores inbound leads and routes the hot ones to an AE.
Saves ~$300–500/mo
Custom
Any workflow. One sentence — or one demo call. WUPHF builds it.
Yours saves the most.
Source available
FAIR-CODE
LICENSE
Free to self-host. Runs on your machine.
The job description is one sentence or one call — not a config format.
No seats. No usage fees. No Michael Scott energy required (but encouraged).
FAQ

Questions the last five skeptics asked.
Answered honestly.

Whatever happened to the "office of AI employees" pitch?
We pivoted. On the internet, in front of everyone. Watching five agents discuss your work turns out to be a form of work — specifically, yours. "AI employee" is a metaphor; an agent that files the report every Monday at 9:00 is a fact. So WUPHF now builds the second thing. Everything your old office wrote — wiki articles, notebooks, HTML and PDF briefs — is preserved verbatim in each agent's Knowledge tab. We kept your homework; we just fired the fictional middle managers.
Will it send an email or a Slack message without asking me?
No. That is the approval gate and it is not optional. Reads are free — the agent can look things up all day. Writes are held: any email, Slack post, or CRM write is drafted and parked until a human taps approve. The run history shows exactly what it wanted to send and when you approved it.
What happens to my old WUPHF wiki and notebooks when I upgrade?
They come along. Team wiki articles, each agent's notebook notes, and the sanitized HTML/PDF briefs the old product generated are all served verbatim as Knowledge pages, labeled "Preserved from your previous workspace," with the visual briefs viewable inline in a sandboxed frame. Nothing is re-synthesized, nothing is lost, and nothing pretends it wrote your notes.
How is this different from Zapier or n8n?
Those are tools for building automations: you draw the flowchart, you own every branch, and maintaining the diagram slowly becomes the job. WUPHF is closer to hiring: you say the job or demo it once, and the agent keeps the branches. It also ships things a flowchart does not have — its own app UI, versioned scheduled routines with run history, self-authored tools, and cited knowledge.
What actually runs 24x7? My laptop sleeps.
Honest answer: agents run as long as the machine running WUPHF is on. On a laptop that means "while the laptop is awake." Put it on a Mac mini in a closet or any always-on box and 24x7 means 24x7. If you would rather not own a closet server, Nex Cloud is the hosted version.
What happens if I have not connected a model or any integrations yet?
The whole pipeline still runs — routines fire, transcripts and run history record — but outcomes are simulated, and they say so in plain text: "Simulated run: no model or integrations are connected on this host." No fake demo data pretending to be your pipeline. Connect a provider and the same routine does the real thing.
Does my data leave the machine?
The runtime is local and your workspace state lives in local files you can cat. Three things do leave, all under your control: inference calls to whatever model provider you configure (point it at a local model and even that stays home), the integration calls you explicitly connect via Composio, and anonymous product analytics — which you can opt out of. We would rather tell you about the analytics here than have you find it in devtools.
Is this production-ready?
It is v0.x and we put that on the website. Some things will break; when they do, per-run transcripts show exactly where, and the issue tracker is public. The catch you were scrolling for is just that it is early. Early, local, and honest beats polished, hosted, and vague — but that is a preference, and it is yours to make.