Running your practice on Karbodh

Karbodh installs as a native Windows application. Before you can use any feature, the app needs to be activated and walked through a one-time setup.

Installing

Run Karbodh_Setup.exe and follow the installer. It installs to your local user profile — no administrator rights are required. A desktop shortcut is optional and offered during setup. Your data (client records, notices, chat history) is stored separately from the program files, so reinstalling or updating Karbodh never touches your existing records.

Activating your license

On first launch, Karbodh shows a full-screen activation card before anything else is usable — this is intentional: the app cannot be opened without a valid license or an active trial.

Enter your firm's name and either start the free trial or type in a license key in the KRBD-XXXX-XXXX-XXXX-XXXX format, then choose Activate License. Your license is bound to this computer's Hardware ID, shown at the bottom of the activation card — support will ask for this if you ever need to transfer or troubleshoot a license.

First-run setup

Once activated, a short setup walks you through configuring the app for your firm:

1. Choose your AI provider. Pick from OpenAI, Anthropic, Gemini, or Karbodh Native — built-in local models that run entirely on this PC with no external API key or internet dependency. Native is the default and is a reasonable choice if you'd rather not send data to a third-party AI provider.
2. Set up users. Choose "Just me" for a single-user install (with an optional instance password), or "My team" to create a named admin account and enable multi-user mode from the start. You can switch to multi-user later from Settings — see Users & roles.
3. Review data handling. A short screen discloses what data each AI provider you might choose receives. Nothing here is a decision you're locked into — it's all reconfigurable later in Settings.
4. Optional survey. A quick "how will you use Karbodh" question, skippable.

After setup, you land on the home chat screen with an empty workspace ready to go. Here's what those four screens look like, in order:

Onboarding: Welcome → LLM Preference → User Setup → Data Handling & Privacy → (optional) Survey.

The Dashboard is the home screen for the Income Tax Notices side of Karbodh — a snapshot of your whole client roster at a glance.

Four tiles summarize the whole practice: Total Clients in your roster, clients with an Active Notice outstanding, Clean clients with nothing pending, and how many notices have been Processed via AI so far. Below that, a recent-activity feed and a client-status breakdown show what's changed since your last sync.

The Portal Sync button here does exactly what it does everywhere else in Karbodh — see Portal sync & notices for the full flow. Two quick-action shortcuts, Add Client and Ask AI, jump straight to the two things you'll do most often.

Your client roster — the PAN, portal credentials, and contact details Karbodh needs to sync notices on each client's behalf.

Adding a client

Choose Add Client and fill in the record. The DOB and Portal Password fields matter more than they look — Karbodh derives the Income Tax portal login and the PDF-decryption password for that client from them automatically.

Importing many clients at once

Choose Import from Excel and upload a spreadsheet with the columns pan, name, dob, portal_password. Karbodh reports how many rows were imported and how many were skipped (usually duplicates or malformed PANs).

Finding a client

Search by PAN or name, or filter by status: All Statuses / With Notices / Clean / Errors / Pending–Never Synced, plus a second filter for incomplete records — Missing Password / Missing Email / Missing Mobile — useful for finding clients you need to follow up with before the next sync.

Syncing

Tick one or more clients and choose Sync Selected, or use the single cloud-upload icon on any row to sync just that client. Either way, this hands off to Portal Sync — covered next.

Portal Sync is Karbodh's automated bot: it logs into the Income Tax e-filing portal on behalf of each client and pulls down their notices and proceedings — the same task you'd otherwise do by hand, client by client.

Running a sync

1. Choose Portal Sync from the Dashboard, Client Management, or the Notice Management page.
2. On Notice Management, choose the scope first — All Clients or Selective Clients (opens a picker with search and Select/Deselect All).
3. Choose Start Sync. Karbodh opens an automated browser window against the e‑filing portal and streams live progress back into the app — a per-client status feed and a progress bar.
4. When it finishes, a summary banner appears, and Windows shows a desktop notification even if Karbodh isn't the focused window.

What gets downloaded, per client: notice PDFs ({PAN}_Notice_P{n}_N{n}.pdf), closure reports where applicable, and a Proceedings Summary spreadsheet used by the Flash Report feature in the next chapter.

Two panels help you keep track over time: Sync History (past sync runs and their reports) and Last Sync Duration (total time and average time per client, useful for planning a large sync).

Tuning sync performance

The Performance & Automation panel (visible on the Notice Management page) lets you match Portal Sync to your PC's capability:

Viewing notices

The Notice Management page groups clients into expandable cards. Each proceeding row shows the Assessment Year, section, a status pill, and the notice/compliance/response dates, split into Active and Historical tabs. Choose the PDF icon on any row to preview the notice or closure report inline — Karbodh decrypts it automatically using the client's PAN and DOB.

Where Karbodh turns a downloaded notice into a drafted reply. Open a client, and four actions cover the whole workflow from raw PDF to a reply ready for your review.

Every AI-drafted reply carries the same footer: "AI responses are advisory only. Always verify with official tax records."

The four actions

Working with a reply

Any AI response can be copied or downloaded as a formatted PDF. Conversations are kept per client as sessions — use Previous Sessions to revisit an earlier draft, or New Session to start clean. Before anything is sent to the AI provider, Karbodh automatically redacts personal identifiers (email, phone, PAN) from the notice text.

PageIndex — vectorless document reasoning

PageIndex is a separate, general-purpose tool for indexing any PDF or Markdown document — the Income Tax Act, a circular, a client agreement — into a hierarchical "reasoning tree" you can then ask questions against, without building a vector database.

1. Choose Index Document, give it a file path and an optional display name.
2. Once indexed, select the document and type a question in plain language.
3. Choose View Tree to see the document's full outline — titles, sections, and AI-written summaries at each level.
4. Every question you've asked is kept in a Query History panel you can revisit.

Alongside the notices tools, Karbodh includes a general-purpose AI chat assistant — its own documents, its own memory, separate from any one client.

Workspaces

A workspace is a named, self-contained AI setup — its own model, its own documents, its own prompt. Create one with + New Workspace. Typical uses: one workspace per practice area, one per recurring task, or one per team member's own scratch space.

Inside a workspace, threads let you branch a conversation without losing the original — fork a thread to try a different angle, or bulk-delete old ones during a clean-up.

Workspaces (left) hold their own threads; the AI's saved memories (right) can be scoped to one workspace or promoted to global.

Documents & memory

Drag a file (or paste a link) into a chat to attach it — Karbodh embeds it so the AI can reference it in that workspace going forward. Separately, the AI can save a memory — something worth remembering across conversations — scoped either to one workspace or promoted to global so it follows you everywhere. Manage saved memories from the Memories panel: edit, delete, or move between workspace-scoped and global.

Chat features

• Streaming responses, with edit / regenerate / pin on any message, and thumbs feedback.
• Text-to-speech playback and speech-to-text input.
• Export a full chat history.
• Per-workspace model override, or hand routing decisions to a Model Router — see Settings reference.

Give the AI a standing instruction and a schedule, and let it run unattended — from Settings → Tools → Scheduled Jobs.

Creating a job

Choose New Job and fill in a Name, the Prompt describing what the AI should do, a Schedule (built visually, no need to hand-write cron), which Tools it's allowed to use, and an Output Format — none, PDF, Excel, or DOCX.

Tools a job can use

Managing jobs

Each row supports Trigger now, an enable/disable toggle, Edit, and Delete. Opening a job shows its Run History — every past run, each with a downloadable output file, a "continue in chat" option to keep working with the result, or a kill switch for a run still in progress. Turn on the notification bell to get a desktop alert when a run finishes.

Karbodh runs single-user by default. Switch on multi-user mode to give each person in your firm their own login.

Turning on multi-user mode

Go to Settings → Security and choose Multi-User Mode. This creates the first admin account and, from that point on, everyone signs in individually rather than sharing one install.

Roles

Adding people

From Settings → Admin → Users, choose Add user and set a username, password, and role. The new account is issued six one-time recovery codes shown once — have the user save them somewhere safe, since Karbodh can't recover a lost password without them (an admin can also reset it directly). You can set a per-user daily message limit and suspend an account from the same screen.

To bring someone on without creating credentials yourself, use Settings → Admin → Invites to generate a link (optionally scoped to a role and specific workspaces). They accept it and set their own password.

By default, every install of Karbodh is standalone — its own data, its own AI, nothing shared. On a Team+ plan, you can instead run one machine as a shared office server that the rest of your firm connects to over your own network.

Found under Settings → Network & Server Connection (admin only), three modes are available:

Setting up the host

On the machine you want to act as the office server, choose Make this the central server. Two more options appear:

• Also run a shared LLM server on this machine — pick one of your already-downloaded native models to serve to every connected seat, instead of each machine running its own copy.
• Also run a shared document search index on this machine — a shared vector database so every seat searches the same document index.

Save your changes and restart Karbodh for the new mode to take effect.

Connecting a seat to the host

On every other machine, choose Connect to another machine and enter the host's address (for example https://192.168.1.20:3001). For an HTTPS address, choose Verify Connection first — Karbodh fetches and pins the host's certificate fingerprint so this seat can confirm it's really talking to your office server, not an impostor on the same network, before you save. Save and restart to connect.

Every settings screen in Karbodh, grouped the way they appear in the app's Settings sidebar — illustrated so you know what you're looking for before you go hunting for it.

AI Providers

Admin

Users and Invites are covered in full in Users & roles. The remaining admin screens:

Customization

Tools

Security

Privacy & data handling

Developer API

About & License

See Plans & licensing — the full activation/status screen is illustrated there.

What your license screen tells you, what each status means, and how seats work once you're running more than one machine or user.

Settings → About & License

This screen shows your Licensee, the Licensor (CA Ronak Patel), the installed Application Version, RAG Database Version, your active License Key, Subscription Type, a License Status badge, Days Left until expiry, and your Hardware Fingerprint (copyable for support). Two actions live here:

License status messages

How seats work

Every plan above Solo is sold with a seat count. A seat is consumed by either a physically activated machine or a named user account on a central server (see Centralized server sync) — both draw from the same pool, counted the same way. A Server-10 plan, for instance, supports any combination of machines and named accounts that adds up to 10 — three activated laptops and seven named accounts on a shared host is exactly as valid as ten separate installs.

Machines and named accounts draw from the same pool — this firm could just as easily run 10 standalone machines, or 1 host + 9 named accounts.

Buying an Addon seat pack adds seats to your existing plan's pool rather than creating a separate license — the extra capacity is usable by a new machine activation or a new named account, identically to your plan's original seats.

If your seat pool is ever exceeded — most commonly after a downgrade — Karbodh doesn't delete anyone. Instead it blocks new logins beyond the cap until your firm's admin brings usage back within the limit.

HWID

Hardware ID — a fingerprint of this specific PC, used to bind a license key to one machine.

Portal Sync

Karbodh's automated bot that logs into the Income Tax e-filing portal and downloads notices/proceedings for your clients.

PAN

Permanent Account Number — the client identifier used throughout Karbodh, in the format AAAAA9999A.

AO

Assessing Officer — the tax authority a notice reply is addressed to.

Flash Report

An AI-generated summary of a client's Proceedings Summary spreadsheet.

PageIndex

Karbodh's vectorless document-reasoning tool — indexes a PDF or Markdown file into a hierarchical outline you can query, without a vector database.

Workspace

A self-contained AI chat setup with its own model, documents, and prompt.

Seat

One unit of your license's capacity — consumed by either an activated machine or a named user account on a shared server.

Central server / host

The one machine, on a Team+ plan, that other seats connect to over the office network instead of running an isolated install.

Model Router

A configurable set of rules that sends a chat request to a different AI model based on conditions you define, with a fallback if none match.