Show your work, not just your conclusion.
You recommend an architecture. The client asks why not the other one. Kirat lets Claude put that question to ChatGPT and Gemini as well. Three model families take positions independently and vote, and the result keeps each position and the objection that lost. You still own the recommendation.
Judgment you can't show is just assertion.
"Why not X?" Your answer is your experience. That's usually enough. Usually.
The recommendation goes out under the agency's name — whoever on the team actually wrote it.
Months later, nobody can reconstruct why the call was made. Including the person who made it.
A transcript, not a verdict.
For a question you send through the council, the result lists each member's position and reasoning, its self-reported confidence, the outcome and how it was reached, and the cost. If there is no majority, Kirat says so (needs_human) instead of forcing agreement. The result documents that the question was examined from more than one direction. It does not prove the answer is right. That part stays yours.
FIG. 1 — Two runs from our own decisions. The second corrected an earlier condition, which rested on incomplete facts Claude had given the council. You can only fix a premise you can see.
The condition from 8 October is why these limits come before any price on this site. When the council later voted to support Claude Code cloud sessions, it set the same condition for the cloud limits. Today Kirat runs in Claude Desktop (macOS/Windows) and Claude Code, and in Claude Code cloud sessions with a one‑time setup — not in chat in the browser or the mobile apps, or in Cowork tasks that run in the cloud.
What this does not do
- —It does not make your recommendation correct. Different model families fail differently, but they can still be confidently wrong together. A 3:0 vote is agreement, not proof, and the confidence values are the models' own self-report.
- —It is not a data-protection measure. Kirat has no server, and we never see your keys or your data. But everything you give it (questions, context, code, images) goes directly to OpenAI and Google, under your accounts and their terms — from your computer, or in Claude Code cloud sessions from your own cloud environment through Anthropic's network proxy for that environment. Minimize or redact client data first, exactly as you should with the AI tools you already use.
- —It is not a deliverable. The result is working material. What the client reads is what you wrote.
- —It is not an archive. Kirat keeps the last 20 results in the
.kiratfolder in your home folder. Save the ones a client may ask about with the project. - —It does not run everywhere you use Claude. Only where Claude runs tools on a machine that has Kirat installed: Claude Desktop chats on macOS and Windows, Claude Code in a terminal or VS Code, local sessions in Claude Desktop's Code tab, SSH sessions — and Claude Code cloud sessions (Code tab cloud sessions, Claude Code on the web or in the mobile app) after a one-time setup of your cloud environment. The cloud is limited: everyone who uses the environment can read its environment variables, including your keys; Kirat's daily budget counts per session, so set spending limits at OpenAI and Google; background jobs and recent results end with the session; and the environment must allow two hosts,
api.openai.comand the Kirat download host. If the setup fails, the session still starts, just without Kirat, and the failed state can stay cached for new sessions until you edit the setup script. Not chat in the browser or the mobile apps, and not Cowork tasks that run in the cloud. WSL sessions and local Cowork tasks are not tested. Networks on your own computer that require an HTTP proxy are not supported. Organizations that restrict Desktop extensions to an allowlist block manual installs; ask your admin. Where it runs, in full.
Before you buy.
Can I use it on client projects under NDA?
That depends on your NDA and on the providers' terms, not on us. What we can tell you precisely: Kirat has no server, no account with us and no telemetry. What you give it goes directly to OpenAI and Google, under your own accounts — from your computer, or in Claude Code cloud sessions from your own cloud environment through Anthropic's network proxy for that environment. Its local data — the daily cost counter, the cost log, the last 20 results (they contain your questions and answers), the log of delegated decisions and generated images — sits in the .kirat folder in your home folder and can be deleted any time.
Compared with using Claude alone, Kirat adds two recipients: OpenAI and Google. If your NDA covers sending this material to them, you are within it. If it doesn't, Kirat doesn't fix that. Keep those projects out of the council.
Can I bill it to clients?
You pay OpenAI and Google directly for what you use, so the API cost is an expense on your own provider accounts like any other. Whether you bill the analysis time is between you and your client. We don't have a "billable AI hours" story to sell you.
One purchase, many clients?
For your own work, yes. A license covers one person, on any number of that person's own devices and own cloud environments, and you can use what you produce with it in client work.
It does not allow redistribution: you can't pass Kirat itself on to a client or a colleague. Each person on your team who runs it needs their own license, and a cloud environment shared with others needs a license for each person. The code is proprietary.
Where do the API keys live?
You enter them in Claude's own install or settings dialog. Claude Desktop stores them in the operating system's secure storage (macOS Keychain, Windows Credential Manager). Claude Code stores them in the macOS Keychain, and on Windows and Linux in ~/.claude/.credentials.json, readable only by your user. In Claude Code cloud sessions, they are environment variables of your cloud environment, readable by everyone who uses it.
We recommend a separate provider project with a spending limit. Kirat's budget guard also blocks calls that could exceed a daily budget ($5 per UTC day by default, adjustable), but it is best-effort — set limits at the providers too.
What if the council is wrong?
Then you have three wrong opinions and a record of them. That is better than one wrong opinion and no record, but it is not a safety net.
Read the dissent first. It is the part a single model reviewing its own answer tends to smooth away. The evidence section on the main page shows a real 2:1 run, dissent included.
One product for everyone: one purchase, one download with the Claude Desktop extension and the Claude Code plugin. Pricing and details are on the main page.