Capabilities
Papi Labs Ltd
Papi Labs is a British engineering company working in applied AI. It builds its own products and takes on other people’s hardest builds.
- Company
- 16762002
- Registered in
- England and Wales
- Contact
- hello@papilabs.ai
What we can be hired to do
- Agentic systems
- Software that carries out multi-step tasks using real tools, with a person reviewing the important parts.
- On-device AI
- Transcription, redaction and other AI that runs on the user’s own machine, for data that cannot leave it.
- Browser extensions
- Built and published in the Chrome Web Store, including in Arabic, Persian and Ukrainian.
- Desktop applications
- Signed desktop apps, sold as a one-time licence.
- Voice agents
- Phone ordering and support in real time, connected to the systems a business already uses.
- Developer tooling
- Tools that help AI coding agents understand a codebase before they change it.
- AI-assisted product design
- From a written brief to a 3D design to a physical product on sale.
- Architecture and review
- A senior review of a system before it is built, and of the code afterwards.
We do not do staffing, long-term managed services, or work that needs clearance above OFFICIAL.
Case studies
Panda
Describe a physical object in plain English and get back accurate design files, ready to print. Panda is the in-house system I use to design printed parts without opening CAD.
- Context
- Papi Labs designs and prints its own physical parts in-house. The company is one engineer, and nobody here draws in CAD.
- Problem
- Learning CAD is slow, and AI shape generators return geometry with no engineering truth behind it. Neither route gives you a file you can trust to print.
- Constraints
- The design and checking tools run on one computer in the workshop.
- Approach
- You describe the part in plain English, and Panda designs it and measures it before anything reaches the printer.
- What was novel
- Nothing is exported until it has been measured and passed. A failure names the part, the measured value and the place on the geometry where it went wrong.
- Outcome
- The 34-part robotic hand passed 326 checks in 50.2 seconds, with none failed and one skipped. Every build leaves a record that can be checked from the first entry to the last.
- Evidence
- The readout on this page is the gate's own report for that build.
- Status
- In development, internal to Papi Labs and proprietary. There is no public release and no published price.
logga
Take meeting notes or dictate, and let AI transcribe and summarise it on your own computer, privately. No subscription, just a one-time fee.
- Context
- Some conversations cannot be sent to a meeting-notes service: client work, legal matters, health, anything covered by an agreement. logga is the recorder for those rooms.
- Problem
- The usual tools upload your audio and process it on someone else's servers. If that is not allowed, you are back to typing notes by hand.
- Constraints
- The transcription and the summary run on the user's own computer, and every outbound call the app can make has to be listed and defensible.
- Approach
- It runs on your own computer: you record a meeting or dictate, and the transcript and the summary come back there.
- What was novel
- The privacy claim is checkable. The app publishes the full list of network calls it can make, so you can hold it to that list.
- Outcome
- The whole path is built and working, from a recording to a searchable transcript with the speakers separated, to a written summary you can act on.
- Evidence
- The list of network calls the app can make is published with it, so the claim can be read before anyone installs anything.
- Status
- In development and launching soon. Sold as a one-time licence rather than a subscription, with a trial that needs no card and no account.
Sous
An AI that answers your restaurant's phone, takes the order and sends it straight to the kitchen printer. Built on FairPlate.
- Context
- FairPlate® is the ordering platform Papi Labs runs for restaurants. Web orders reach the kitchen printer on their own. Phone orders are still written down by hand.
- Problem
- A ringing phone pulls someone off the line during service, and what they scribble never reaches the system the rest of the orders live in.
- Constraints
- No new hardware in the kitchen, no change to the way a ticket prints, and a size the caller states is never guessed.
- Approach
- Sous answers the restaurant's phone, takes the order in conversation and puts it into the ordering platform the kitchen already uses.
- What was novel
- Matching what a caller says to a live menu is the part that usually fails. Sous checks the order against that restaurant's real menu and asks again rather than guessing.
- Outcome
- Phone orders reach the kitchen the same way web orders do, and nobody has to leave the line to answer the phone.
- Evidence
- A real phone order has run the whole way through, from a customer calling the number to a ticket printing in the kitchen with nobody touching it.
- Status
- In development, running as a supervised pilot.
GuideKit
Learn AI tools while you use them, with coaching in a panel beside your work. In English, Arabic, Persian and Ukrainian. Our AI workshops for teams run under GuideKit too.
- Context
- People are told to use AI at work and left to work it out inside the tool. Papi Labs runs GuideKit as its teaching product, and I run my AI workshops for teams under the same name.
- Problem
- Training sits on a separate platform, away from the tool and away from the work, so what someone can actually do afterwards is never visible.
- Constraints
- Four languages. Arabic and Persian read right to left, so the layout has to change too.
- Approach
- A panel sits beside the AI tool the person is already using and coaches them there, while they do the real work.
- What was novel
- The lesson happens where the work happens. You apply what you have just read in the same gesture, in the tool you were going to use anyway.
- Outcome
- Live at guidekit.ai, with more than 450 installs.
- Evidence
- Anyone can install it and use it today, and a certificate it issues can be checked by whoever is shown it, on a public page.
- Status
- Live. The next version, which lets an organisation show what its people can actually do, is in development.
samnexus
Gives your AI coding agents a full map of your codebase, so they know what a change will break before they make it.
- Context
- Papi Labs is one engineer and a large number of AI agents, working across several live codebases at once.
- Problem
- An agent that cannot see who calls a function will still edit it confidently. The damage shows up later, in a file nobody opened.
- Constraints
- It runs on your machine. Your code never leaves it, and every answer carries a label saying how far it can be trusted.
- Approach
- It builds a map of your codebase on your own machine and lets the coding agent consult that map before it edits anything.
- What was novel
- Before a change, the agent gets one briefing: what the code is, what uses it, what depends on it, what tests cover it, and how risky the change looks.
- Outcome
- Agents stop making confident edits to code that something else was quietly relying on, and the person supervising can see why a change was judged safe.
- Evidence
- Papi Labs builds all of its products with it, on live codebases, every working day.
- Status
- Live and in daily use inside Papi Labs, on several codebases at once.
Security and privacy posture
- logga runs its transcription and summaries on your own computer, and publishes the full list of network calls the app can make.
- samnexus runs on your machine and makes no network calls.
- ClipShield matches patterns in your browser. The original values are held in memory for the session and are not written to disk.
- Panda’s own toolchain runs locally.
- Where a product uses hosted services, they are listed in that product’s own documentation. We can send the list on request.
- This website makes no third-party requests. It has no analytics and sets no cookies.
- Registered with the Information Commissioner's Office.
Working with us
We take on a few outside projects each year. If you have one in mind, write to us.
- Email us about what you want to build.
- We have a half-hour call. If it is not a fit, we will say so on the call.
- You get a short written scope: what we will build, what it costs and what it depends on.
- Work starts within weeks. You see something running in the first week.