¶ 1 A useful first run

Complete one small, checkable job.

Connect a model. Open Settings → AI connection. Add a direct provider, a router, or local Ollama, then use Test and Refresh models.
Create a matter. Give the deal or dispute a name, client, governing law, and its other jurisdictions.
Choose its folder. NyayVaani reads only inside the folder you approve. Attach the PDFs and DOCX files needed for this job.
Start in Work. Pick the matter, choose review or drafting, confirm the source documents, and state the task.
Judge the evidence. Open a citation and compare the answer with the highlighted source before relying on it.

Need installation help first? Follow Getting started.

¶ 2 The starting desk

Work begins the task; it does not own the file.

Use Work when you know the legal job you need done. Choose review or drafting, select a matter, apply a playbook if useful, and confirm exactly which documents are in scope. A review becomes a saved comparison; a draft becomes a durable document.

Work also shows saved reviews and drafts for the selected matter. You can leave a running job and resume it later without rebuilding its questions or source scope.

¶ 3 The durable case file

A matter keeps context together.

Overview: The matter’s health: source readiness, flagged citations, governing law, recent activity, and the next sensible action.
Sources: The documents NyayVaani may use. A folder boundary alone does not attach every file; you still choose the documents.
Work products: Drafts, document reviews, and the matter conversation. Return here to resume work in its original context.

Closing a matter removes it from active work without deleting its history. Removing a matter from the app does not delete the files in its folder.

¶ 4 The evidence boundary

Only attached, ready sources ground the answer.

Choose a trusted folder, then attach the documents required for the matter. NyayVaani extracts text and records source anchors so answers can lead back to the relevant page and passage. A scanned document that is read through OCR is labelled as such; a file that cannot provide usable text needs attention before it can support grounded work.

Scope rule

Before review, chat, or drafting, read the reference-document list. Checked documents may be sent to the selected model connection; unchecked documents are excluded from that request.

¶ 5 What remains

Work products are saved, resumable outputs.

A review asks the same questions across selected documents and preserves the resulting table. A conversation keeps follow-up questions with the matter. A draft preserves its title, brief, sections, sources, progress, and generated text.

Needs attention does not always mean the whole job failed. Open the item and inspect failed rows, missing evidence, or a source that could not be read. Retry only after the underlying connection or source problem is corrected.

¶ 6 Repeatable judgment

A playbook is what you ask repeatedly, kept once.

A prompt playbook can start a conversation or supply a drafting brief. A column playbook starts a document comparison, asking each saved question of every selected document. Keep names concrete enough that another lawyer can choose the right playbook months later.

A playbook supplies instructions; it does not silently expand the source scope. Confirm the matter and documents each time you run one.

¶ 7 Claims you can interrogate

A citation is a route back to the record.

Verified in source: The cited words were matched against the current extracted text. Open the passage and judge its meaning yourself.
Matched against OCR text: The words were found in text read from a scan. Check the page image because OCR can misread characters.
Not found: The review found no support in this document. It does not prove the proposition is absent elsewhere.
Unverified or flagged: The passage could not be matched. Treat the output as a lead, not as verified evidence.

Opening a citation takes you to the source viewer and highlighted passage. Use the viewer’s back control to return to the originating review or conversation; browser history may return to a different app state after a restart.

¶ 8 Choose the boundary

The connection determines where selected text goes.

On this computer: Ollama runs the chosen model locally. NyayVaani does not install, start, or download Ollama models for you.
Direct provider: Anthropic, Gemini, OpenAI, Fireworks, xAI, Groq, or Together AI receives selected text directly under its own terms.
Model router: OpenRouter receives the request and may pass it to a downstream provider. Check both boundaries before using confidential material.

Provider keys are checked and stored in this computer’s operating-system keychain. They do not pass through NyayVaani. Under Advanced model routing, you can choose different connections and models for chat, review, drafting, and other jobs.

¶ 9 When nothing moves

Check the boundary closest to the failure.

No model is available: Open Settings → AI connection. Test the connection, refresh its models, and confirm Advanced routing points the job to an available model.
Ollama is offline: Start Ollama at localhost:11434 and install at least one model. NyayVaani will not start it or download a model.
The provider rejects the request: Confirm the key, account credit, model ID, and provider availability. Environment-supplied keys take priority over keys saved in the app.
No documents appear: Choose the matter’s folder, attach the files, and wait until extraction reports that they are ready.
A citation will not open: Confirm the source is still present inside the trusted folder. Resume the saved work product and try the citation again.
Generation failed or stalled: Your brief and saved work remain local. Correct the connection or source problem, reopen the work product, and retry.

Still blocked? Email hello@nyayvaani.ai with the action, exact error text, operating system, and app version. Do not send confidential documents by email.

Know where the work lives.