Frequently asked questions
Every question we get asked, answered plainly.
36 answers on what we build, how we work, what it costs, and what happens to your data. Each one starts with a direct sentence you can quote.
Last reviewed
About ByteKnot Labs
What does ByteKnot Labs do?
ByteKnot Labs is an AI-native product studio and full-stack Java engineering team, founded in 2024. Half of its work is AI-native: assistants, agents, AI-native CRMs and custom models, trained, tested, validated and deployed on the client's own infrastructure. The other half is traditional full-stack Java: CRMs, ERPs, portals and APIs on Spring Boot and MySQL. It also ships four products of its own: ByteMirror, ByteLoader, ByteKeys and ByteViewer.
What does "AI-native" mean at ByteKnot Labs?
It means language models are part of both what we build and how we build it. AI products are designed around what a model can be trusted with and measured with an evaluation set before launch. Our own engineering uses AI coding assistants, with every change read and signed off by an engineer.
Who are your clients?
Founders who need a first production release, product teams adding AI to software they already sell, and companies replacing or modernising a CRM, ERP or back-office system. We work remotely with clients worldwide.
Has ByteKnot Labs shipped anything we can try?
Yes, four products. ByteMirror brings phones into OBS Studio on Windows, ByteLoader sideloads and renews iPhone apps from Windows for free, ByteKeys is an open-source control centre for Linux laptops, and ByteViewer is a native iOS app with its source on GitHub.
Is ByteKnot Labs a registered company?
Yes. ByteKnot Labs is the trading name of ByteKnot Labs LLP.
AI products, agents and models
What is an AI-native CRM?
An AI-native CRM is a CRM designed with the model inside it from the start. It reads call notes and email, drafts follow-ups, scores leads against your own history, summarises accounts and flags exceptions for a person to decide. It still has contacts, pipeline, permissions, reports and integrations.
What does train, test, validate and deploy involve for a custom AI model?
Train means fine-tuning a model or building a retrieval index on your data, in your account. Test means running a held-out evaluation set of real cases on every change. Validate means guardrails, red-team checks, cost and latency budgets and a human sign-off. Deploy means serving it on your cloud or hardware with tracing, monitoring, cost limits and a rollback path.
How much does an AI product or agent cost?
It depends on how many systems it touches, whether it drafts or acts on its own, the state of your data, how much a wrong answer costs and where the model runs. We give a fixed price once the scope is clear, or bill two-week sprints while it is still being discovered, and every proposal states the number before work begins. Running costs, mainly model usage and hosting, are estimated per request in design and watched in production.
Do we need an AI agent, or would a workflow do?
Often a workflow is enough. If the steps are the same every time, a fixed workflow with a model at one or two points is cheaper to run and easier to test. An agent makes sense when the path changes from case to case and the model has to choose the next tool. We say which one the job needs before we quote.
Will OpenAI, Anthropic or Google train on our data?
Not by default. OpenAI, Anthropic and Google Cloud each state that business and API data is not used to train their models unless you opt in or give permission. We use only the providers you approve in writing, and serve an open-weight model in your network when no outside provider is acceptable.
Should we use RAG or fine-tuning?
Use RAG when the model needs facts that change or must be cited, and fine-tuning when it needs a consistent style, format or behaviour at volume. Many products use both. We decide on your evaluation set, not on a hunch.
How do you stop an AI product from hallucinating?
Nobody can guarantee a model never makes something up, so we design for it: answers grounded in retrieval with citations, constrained outputs where format matters, an evaluation set built from real questions, and human approval wherever a wrong answer is expensive.
Which AI models do you use?
Whichever scores best on your evaluation set at a cost you accept. We usually benchmark Claude, Gemini, OpenAI models and open-weight models such as Llama, Mistral, Qwen and Gemma, and design so the model can be swapped later.
Can AI run inside our own network?
Yes. Open-weight models can be served on your own GPUs or private cloud with vLLM, so no data leaves your network. Hosted models can also run inside your own Google Cloud or Oracle OCI account.
Does ByteKnot Labs build its own AI models?
Yes. Our AI Lab trains and publishes its own models, starting with AksharSTT, an open speech-to-text model for Hindi, English and Hinglish that runs offline on a CPU. Its weights, training records and scores are public, so you can check the work before you hire us for a custom model.
Can you add Hindi or Hinglish speech to text to our product?
Yes. AksharSTT turns Hindi, English and Hinglish speech into text offline on an ordinary CPU, from a single 29.5 MB file, so it can run on the device or on your own server with no per-request fee. When your audio is specialised, such as calls, clinics or field recordings, we train it further on your recordings in your own account and measure it on an evaluation set built from them.
CRMs, ERPs and Java platforms
Do you build traditional CRMs and business platforms without AI?
Yes. Half of our work is traditional full-stack Java: custom CRMs, ERPs, billing, HR and back-office systems and customer portals, on Java LTS, Spring Boot and MySQL with a React or Next.js front end.
Why Java and Spring Boot for business platforms?
Because a CRM or an ERP has to run for years, under real data volumes, and be maintained by whoever comes after us. Java and Spring Boot give strong typing, a mature security model and predictable performance, and the pool of engineers who can extend them is large.
Should we build a custom CRM or use Salesforce or HubSpot?
Use Salesforce or HubSpot if your process is close to standard and your team is small. Build when your process no longer fits, when per-seat licences across a large team have passed the cost of building, or when the CRM is part of the product you sell.
Can you add AI to our existing CRM or ERP?
Usually, yes. We review the system, add tests around the parts we touch, and add the feature, such as lead scoring, summaries, drafting or an assistant, with its own evaluation set. A platform on Spring Boot and MySQL needs no rewrite.
Can you modernise our legacy Java application?
Yes, incrementally. We add tests around a module, move it to Java 21 or 25 and Spring Boot 4, and repeat, so every step is safe to release and there is no big-bang rewrite. Spring Boot 3.5 left open-source support in June 2026, so many teams are planning this now.
Working together
How does ByteKnot Labs price its work?
In three ways: a fixed price for a fixed scope, two-week sprints billed per sprint for work still being discovered, and a monthly retainer for ongoing operations. Every proposal states the number before work begins. There is no published price list.
How long does a project take?
Most first production releases ship in six to twelve weeks. Scope, success criteria and the date are agreed before work starts, and you see a working build every week.
What happens after launch?
You can run it yourself with the runbook and documentation we hand over, or keep us on a monthly retainer for monitoring, upgrades, incident response and new features.
Do you sign NDAs?
Yes, yours or ours, before we see anything sensitive.
What time zones do you work in?
We work remotely and agree overlap hours at the start of each project, with a weekly demo and a shared channel for everything in between.
Data, security and ownership
Where does our data live, and does any of it train a model?
Your data lives in your own Google Cloud or Oracle OCI tenancy and your own GitHub organisation. It never trains a model. Fine-tuning happens only when you ask for it, on your data, in your account.
Who owns the code and the models?
You do. Repositories, infrastructure code, prompts, evaluation sets and model weights are yours, with no licence back to ByteKnot Labs.
What happens if we stop working with you?
Nothing has to move. Code is in your repositories, infrastructure is in your cloud as Terraform, and we hand over documentation and a runbook. Our access is revoked and you keep everything.
Which AI providers will see our data?
Only the ones you approve in writing, under their API data terms. Where that is not acceptable, we serve open-weight models inside your network.
Do you comply with India's DPDP Act?
We work to it today, ahead of 13 May 2027, when most of the DPDP Rules apply. This website already meets the notice, consent and withdrawal requirements: analytics runs only if you accept, and privacy requests are answered within 30 days. On client projects we act as your Data Processor under a written contract, and we can build notices, consent records, data principal requests, retention and breach alerts into the system. The detail is in our DPDP compliance policy.
Our products
What is ByteLoader?
ByteLoader is a free Windows app that signs IPA files with your own Apple ID and installs them on an iPhone or iPad, then renews them automatically before the 7-day free signature expires. No jailbreak and no paid developer account.
What is ByteMirror?
ByteMirror is a Windows app that adds iPhone and Android screens and cameras to OBS Studio as sources over USB or Wi-Fi, without a capture card. The free plan runs at 720p30; Pro adds 1080p60. ByteMirror 1.0 is in beta.
What is ByteKeys?
ByteKeys is a free, open-source control centre for Linux laptops. It handles keyboard lighting, Silent, Balanced and Performance modes, fan curves on supported ASUS laptops and a battery charge limit: the controls that vendor apps such as Armoury Crate only offer on Windows.
What is ByteViewer?
ByteViewer is a free iPhone and iPad app that opens saved websites full screen like separate apps, with per-site kiosk mode, Face ID lock and rotation lock, a widget and a Shortcuts action. Its source is on GitHub.
Where do I get support for a product?
Each product site has a guide and a contact route. For anything else, email [email protected].
Taking on new projects
Question not here?
Ask us directly. You will get a reply from an engineer, not a sales script.