AI Answer reads your documents and replies directly. Search only returns the matching passages with no AI and no cost.

Paste text

Index from your storage (URL) - The file is never stored on our servers

Paste a pre-signed / shareable link (Amazon S3, Azure Blob, Google Cloud Storage, or any HTTPS file). CheriMind fetches it into memory, indexes it, and forgets it - nothing is written to our disk. Uses the Namespace/Category chosen above.

All documents

Orchestration stages

Chat

Budget burn-down

Estimate a question's cost before you ask it

Two methods, picked automatically. With enough history: predicted from your most similar past questions (embeddings + nearest-neighbour), and it shows you those questions. Without history: priced from the prompt size and your model's rate card. The result says which was used - see Help for the detail.

Cost by category - Which parts of your knowledge base drive spend

Daily spend (last 30 days)

Recent requests

When you use your own LLM key/resource, these numbers reconcile 1:1 with your provider's billing dashboard.

How to pay

After your free period, pay using any of the details below and keep the transaction reference (UTR). Then let us know, or we will confirm it against our records.

Your plan

Your models - Add several; one is the primary that answers

Configure one or more LLMs. The primary (default) model answers every question; switch it anytime with Make primary.

Document storage - Where your document vectors live

Keep your documents in our managed store (isolated to your project), or bring your own database so they never touch our system.

Live data - Answer from current data in your API or database

By default answers come from your indexed documents. You can also pull current data at question time from your own API or database - useful for live values (balances, order status, inventory) that would be stale if indexed.

Memory - Remember context to improve answers

Answers can use remembered context: session (this conversation), user (this person across visits), and enterprise (organization-wide facts). Memory is context only - it is never cited and never leaves your project.

What is CheriMind?

CheriMind turns your own documents into a trustworthy AI assistant that can run entirely inside your own network, so your most confidential files never leave the building. Upload your files (policies, FAQs, contracts, manuals) and anyone on your team can ask questions in plain language and get clear answers built only from your content, with the exact sources shown for every answer. No coding, no data science, no guessing.

Think of it as the private alternative to cloud AI assistants like Gemini and Copilot: the same kind of grounded, cited answers, but your data never has to go to someone else's cloud. That is what makes it usable for regulated and confidential work where those tools cannot go.

• Grounded & cited by design - a glass box, not a black box. Every answer is traceable to your sources.
• Your data stays yours - a four-rung trust ladder, from our managed store all the way to fully self-hosted where your files and database never touch us (see below).
• Your AI key, your rules - four key-handling modes, from fully managed to "we never see your key or your answer".
• Provenance and audit built in - every stored piece is stamped with its source, a checksum, a timestamp and who wrote it, so you can prove exactly what is in your knowledge base.
• Live data, not just documents - answer from current data in your own API or database at question time (balances, order status, inventory), on its own or fused with your indexed documents.
• Cost Intelligence - spend forecasts, per-question cost prediction from your own history, and spike alerts - rare in this category.
• Any document, any domain - PDF, Word, Excel, images (OCR) and more, for any industry.
• Zero-code onboarding - add a new team or client as a project - no rebuild, no developers.
• Run it on your own premises - the entire platform (app, database, and AI/embedding service) can be deployed inside your own network or data centre, so nothing ever leaves your walls.

You decide how much stays with us. Start on any rung and move anytime - your projects and users are unaffected.

On-premises deployment. For the strictest requirements, CheriMind can be installed entirely on your own servers - the .NET application, the PostgreSQL database, and the Python AI/embedding service. Nothing leaves your walls, so it suits regulated, private, or air-gapped environments. You keep full control and we operate none of it.

Configure rungs 1-4 in Settings -> Document storage. Full on-premises hosting is a setup option - raise a Support ticket and we will help you deploy it.

Open the round ? button on any page for a detailed, page-specific guide with examples.

CheriMind can answer from three kinds of source, on their own or combined: your documents, a database, and an API. Here is exactly what happens in each, and how a question becomes an answer.

1. Documents (indexed knowledge)

Use for content that does not change minute to minute: policies, manuals, FAQs, contracts.

When you add a document (Indexing page): CheriMind reads the file and pulls out its text (PDF, Word, Excel, CSV, HTML, or even a scanned image via OCR). The text is split into small overlapping chunks of a few sentences each. Every chunk is turned into an embedding - a list of 384 numbers that captures its meaning, so pieces about the same topic sit near each other in "meaning space". The chunks and their embeddings are stored; the original file is discarded, and each chunk is stamped with its source, a checksum, a timestamp, and who wrote it.

When you ask: your question is embedded the same way, and CheriMind finds the chunks whose meaning is closest to the question (semantic search, not keyword matching), and hands the top matches to the answer step.

Configure: Indexing page - upload a file, paste text, or index from a URL. Where the vectors live (our store or your own database) is set in Settings -> Document storage.

2. Database (live SQL)

Use for current values that would be wrong if frozen in an index: a balance, order status, stock level.

At ask time, CheriMind connects to your database and runs a query you wrote. Your query contains one placeholder, @p; CheriMind binds the lookup value to it (never string-glues it, so it is safe from SQL injection), runs the query, and turns each returned row into a piece of context for the answer. Nothing is stored - it is read fresh every time.

Example query: SELECT name, balance, status FROM accounts WHERE customer_id = @p

Configure: Settings -> Live data -> Database (SQL): enter host / port / database / user / password, write the query, pick how @p is filled (see "What fills @p" below), Test, Save. The connection is encrypted; the host cannot point at an internal address.

3. API (live REST)

Use when your current data already sits behind an HTTP service.

At ask time, CheriMind sends a POST to your endpoint with a small JSON body - {"question": "...", "key": "...", "context": {...}} - and reads the JSON you return (either a plain array of records, or an object with a results/data array). Each record becomes a piece of context. Add an Authorization header if your API needs one.

Configure: Settings -> Live data -> REST API: enter the endpoint URL and optional auth header, pick what fills key, Test, Save. The endpoint is encrypted and checked so it cannot reach internal addresses.

What fills @p / key (the lookup)

• By a context field (e.g. customerId): the value comes from the User context box on Ask (or the userContext your app sends). So {"customerId":"C-102"} fetches exactly that record. Best for "my ..." questions.
• By the question text: the question itself is passed, and your query/API decides what is relevant. Best for search-style lookups.

Which sources are used (the mode)

In Settings -> Live data you choose Documents only (default), Live only, or Both. In "Both", documents and live data are searched together and merged, so one answer can carry the background (from documents) and the current value (from live).

4. How Ask turns that into an answer

Every question runs through the same pipeline (you can watch it live on the Pipeline page):

1. Plan - decide which sources to use, based on your mode and the question.
2. Retrieve - search documents by meaning, and/or run your database query / call your API, all at once.
3. Fuse - merge and rank everything that came back, keeping the most relevant pieces.
4. Govern - apply privacy rules (e.g. hide emails / phone numbers) before the model sees anything.
5. Prompt - assemble a grounded prompt: the question plus only the retrieved context.
6. Answer - your chosen AI model writes the answer from that context. You get the answer, a grounding badge (was it backed by your sources?), and citations - the exact pieces it used, whether a document chunk, a database row, or an API record.

If nothing relevant is found, CheriMind says so rather than guessing. The AI model itself is set in Settings -> AI model, and how much of your key we hold is up to you (four key-handling modes).

1. Indexing - upload your documents (or paste text).
2. Settings -> Live data - optionally connect a database or API for current data.
3. Ask - type a question; get a grounded, cited answer.
4. Settings - connect your own AI model and choose where your data is stored.
5. Scorecard & Cost Intelligence - track answer quality, coverage gaps and spend.

Tip: the round ? button (bottom-right) gives help specific to whichever page you're on.