Live
Live API-to-CRM integration showpiece
This is a live run: public feeds move through four timed stages into a small tool that keeps leads. [SQLite]
Custom work, built and tested, every claim with a dated receipt.
For the person reviewing a proposal: every service, the numbers behind it with their sources, a live count from this server and how a project runs.
- median run, last 33 recorded runs
- 1.63 s
- fresh locked-down container per run
- 1
A fresh locked-down container per press: real weather data in, a go / watch / reschedule call out.
Checked Open the live runner
What we build
We build websites and apps your team and customers use, and tools that collect and tidy up information. Others are a chat on your website and a phone receptionist. The last is practice attacks that find weak spots in your bot.
- Web apps and websites. We build websites, internal tools and phone-first apps, from the database to the screen. [Astro 6] Eight example websites hold 309 pages. See the web app work.
- Scraping and data tools. Public data collected, cleaned, matched across sources and handed over ready to use. [Python and DuckDB, written matching rules] 3,361,791 raw county records became 1,391,184 parcel records. See the data work.
- Chat bots. A chat bubble on your site that greets visitors in your brand colors and emails each lead to your team. [TypeScript widget, 46 industry playbooks, WCAG contrast check] See the chat work.
- Voice bots. A phone receptionist that says it is an AI, finds out what the caller needs and looks things up while they wait. [xAI voice console, web search and end-call tools] We have built voice bots that take real calls. See the voice work.
- Red-team testing. We play tricky callers and chat users against your bot, on purpose, and check what it says and does. [7 attack groups, fixed-rule checks, no AI judge] The record: 45 test conversations, 40 of them tricky and 5 everyday, with zero failed checks on the final run. See the testing work.
- Integrations and automation. We connect your customer tracker, phone line and website, so you stop being the go-between. [public API pull, SQLite, signed webhook demo] A public feed became records in a small CRM in four timed stages. See the service menu.
Receipts table
Live
Live code runner
This is a live reading: the fixed weather script reads the real forecast and shows its log as it runs. The last 33 recorded runs had a median of 1.63 s.
- 11/11 Weather runner sandbox checks passed Build notes Checked
Replay
Voicebot red-team suite: replay of the final full run
A real run, replayed: 45 conversations played twice made 90 runs, and the final run had zero failed checks. [grok-4.7]
See it work
Each one is a complete business website, with its pages, forms and checks in place. Browse the site work.
From nine county sources, 3,361,791 raw county rows became 1,391,184 land records. The tool marks repeated records without deleting them and blocks import if the do-not-call list is missing. Follow the data work.
Visitors get starter buttons suited to the trade, then a card to confirm their details. The bubble emails the lead summary. See the chat work.
Replay
Astro site factory: eight trade-business sites, research and checks
A real run, replayed: eight business sites cover 309 pages, built through 12 build stages from 27 to 30 Sep 2026. Checks compared all 28 site pairs for similarity.
Run Open
Replay
Lead scraping pipeline: eight-stage replay
A real run, replayed: the eight-stage call-list pipeline, played back stage by stage.
Replay
Chat widget on a water-damage company's site
The chat bubble carries 46 trade playbooks in one widget of about 100 KB that loads no outside libraries. [zero runtime dependencies]
Own code and how it is checked
Written here
We direct AI coding agents and check their work with tests and gates before anything ships.
The data tools clean and match public records, and the chat bubble greets visitors in your colors and emails lead summaries. The phone receptionist opens each call by saying it is an AI. See the stack page for the open-source parts.
Checked every time
Checks catch type errors before anyone shares code. They also stop one part from reaching into another's internals. [pre-push gate, boundary lint]
Bots review their own draft
Your website text becomes draft settings for the bot. The bot reviews the draft, and rule checks run before a separate step activates it.
Tested by attack, on purpose
Test callers press on the receptionist's written instructions, and fixed rules score every reply.
The checks are checked too
Across 36 test transcripts, a self-test flagged every bad one and passed every good one.
Backups restored
In the documented restore drill, five databases came back with matching row counts.
How a project runs
- Send what you want built and the tools you already use. We reply with questions or a scope.
- Start with a small first step, such as an audit or pilot. That gives you something concrete to review.
- Agree what the build will cover. Then review small steps as we build.
- You open the real thing and point out changes. We adjust it with you.
Track record
We have converted sites to and from WordPress. See a conversion
We have built voice bots that take real calls. Hear a call
We have moved sites for clients. See a move
We have delivered this work for clients. Read the build records
Why work with Nick Rue
One builder handles your project from start to finish. It is never handed off or put in a ticket queue.
Your messages go straight to us. You can discuss the next change with the person building it.
Before sending a brief, read about the builder or browse the tools used.
Tell us what you would change
One sentence about your project is enough to begin. Tell us what you would change, so a custom version can start with your tools.