Sovereign AI for CA practices
AI that reads your documents — without sending them abroad.
Lipik is FirmsNest’s sovereign AI assistant. It runs in India on Sarvam, never trains on your data, and reads your firm’s documents to pull out figures and draft working papers for your review — unlike assistants that process client data on a foreign model.
Runs in India Never trained on your data You review, compute and file
Why sovereignty matters
Most AI tools trade useful answers for sending data abroad.
For a CA firm holding PAN, Aadhaar and returns, where the model runs is not a detail — it is the difference between an assistant you can use under the DPDP Act and one that quietly ships client data out of the country. Lipik is built to avoid that trade — four guarantees make it hold.
Powered by Sarvam — sovereign AI, built and hosted in India
The model runs in India
The AI that reads your documents runs on Sarvam — sovereign AI built and hosted in India — under terms that keep processing in-country. Nothing is shipped abroad to be read.
Never trained on your data
Your firm’s documents are not used to train, tune or improve any model. What you put in answers your question and nothing else — no retention for model-building.
Documents and index stay sealed
The private search index and the documents themselves sit in encrypted storage in India, isolated to your firm alone — one firm’s data is never visible to another.
Web search is scrubbed
When Lipik checks a public rule, only a generic question leaves; a guard removes anything resembling a PAN, Aadhaar, email or phone number before a search goes out.
What Lipik does
It reads, extracts and drafts. You decide.
Sovereignty is not a compromise on usefulness. Lipik takes the reading and re-keying off your plate and hands you drafts with their sources attached — the work a diligent junior would do before you pick it up.
Reads your documents
Ask about a client’s file in plain language. Lipik answers from what your firm has uploaded, and cites the documents it used so you can verify in seconds.
Pulls the figures out
Point it at a Form 16, a 26AS or a set of accounts and it extracts the values into a structured form — so you review numbers instead of re-keying them.
Drafts a working paper
It assembles a draft tax-estimate worksheet, labelled a draft for your review. The arithmetic is done in code, not guessed by the model.
The honest contrast
Not every AI assistant is built the same way.
Plenty of practice tools now ship an AI feature. The question worth asking is where the model runs and whether your client data is used to train it.
Lipik — sovereign, in India
- Runs in India on Sarvam
- Never trained on your data; no retention for model-building
- Reads your documents, extracts figures and drafts working papers
- Identifiers stripped before any public web search
Assistants on a foreign model
Some practice tools build their AI on a model hosted outside India. CA Manage, for instance, markets an assistant “powered by Gemini”, which means client data is processed by a non-India model. That can be a fine product in other respects — but for a firm holding PAN, Aadhaar and returns, it places client data on infrastructure beyond India.
Competitor positioning is described from their own public marketing, accurate as of our last review.
What stays your job
Advisory, not autonomous.
Lipik does not compute or file returns, and it is not a tax, audit or legal adviser. You continue to compute and file wherever you do today; every figure it surfaces is a draft for you to verify, and every worksheet says so.
We named it after the munshi — the scribe — on purpose. A scribe prepares the record; the principal reads it and signs. That is the relationship we want you to have with it: it prepares, you decide.
Questions
What firms ask about sovereign AI.
What does “sovereign AI” mean for a CA practice?
It means the AI that reads your clients’ documents runs on infrastructure inside India, under terms that keep processing in-country and do not use your data to train a model. For a CA firm holding PAN, Aadhaar and returns, that is the difference between an assistant you can use under the DPDP Act and one that quietly sends client data abroad.
Where does Lipik process my clients’ data?
In India. Lipik runs on Sarvam, a model built and hosted in an India region. The documents and their search index are stored in India, and embeddings are generated in-region. The only outbound request is a web search for a public rule, carrying a generic question with all client identifiers stripped out.
How is this different from an assistant powered by a foreign model?
Some practice tools build their AI on a foreign model — CA Manage, for example, markets an assistant “powered by Gemini”, which means client data is processed by a non-India model. Lipik instead runs on an in-India model under no-training, no-retention terms, so client data stays in India and is never used to improve a model.
Is my firm’s data used to train the AI?
No. Lipik runs under no-training, no-retention terms. Your documents are used only to answer the question you ask, and are not used to train, tune or improve any model.
Does Lipik compute or file tax returns?
No. Lipik is advisory. It drafts working papers and surfaces figures for your review; you continue to compute and file in your own software. Tax arithmetic in its drafts is done in code, not by the language model — the judgement and the responsibility stay yours.
See sovereign AI inside your own branded portal.
Book a demo and we’ll show you Lipik read a document, pull the figures and draft a worksheet — without anything leaving India.