Get the argument you don't have.
Working alone means nobody tells you no. Kirat lets Claude put your decision to ChatGPT and Gemini. Each takes a position independently, Claude adds its own, and the objection stays in the result instead of being smoothed away.
It is not a team. It's a second and third opinion from other model families, inside Claude, whenever you ask for one.
The echo is the failure mode.
You decide late at night. By morning it's in production, and it was never once challenged.
Asking a single model gives you one perspective with no independent comparison. Self-review flatters, and you never see which objection quietly got smoothed away.
You quote a job in a stack you only partly know. Nobody checks the estimate before it turns into a commitment with a date on it.
There is nobody to escalate to. When the call turns out wrong, it is wrong with your name on it, and the client has only you to hear it from.
Months later: "Why did I build it this way?" The reasoning is gone, and you rebuild the argument from scratch.
The objection stops being skippable.
Here is a real run from August 2026, before Kirat was an extension. We asked our own council whether to sell an earlier form of this product — a build manual — and if so, how. Two members said yes. One said don't.
FIG. 1 — The real transcript, about the old build-manual plan. The numbers are self-reported confidence.
The dissent didn't stop the project. The plan later changed, shaped by that objection. What you can buy today is neither the majority's plan nor the dissenter's: a ready-to-install extension, bought once. The council doesn't decide for you. It makes the objection impossible to skip.
Is this a replacement for colleagues? No.
People ask whether this replaces having colleagues. It doesn't, and the difference is worth being precise about. A colleague carries context you never explained, notices the problem you didn't raise, and shares the consequences of being wrong. Kirat does none of those things.
What it does is narrower: for a question you ask, you get independent positions from model families that fail differently from Claude — and the disagreement is kept instead of averaged into a confident-sounding answer. That is a structured second look on demand, with a written result. It is not a team.
What this does not do
- —It is not a co-founder, a colleague or a team. The members see the question and the context Claude sends with it — nothing else. No initiative, no stake in the outcome. You bring the question; it brings positions on that question and nothing beyond it.
- —It is not market validation. Three models liking your idea is three opinions. It is not a customer, and it is not revenue.
- —It is not free to run. Every council decision spends real API tokens — $0.02 to $0.15 in our measured runs, depending on effort tier and input size — billed to you by OpenAI and Google. Three models do not cost like one, and you need API accounts with billing at both providers.
- —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.
What does setup involve?
Claude Desktop: unzip the download, open Settings → Extensions → Advanced settings → Install Extension…, pick the .mcpb file and enter your two keys in the dialog. Desktop warns that extensions installed from a file are not verified by Anthropic; that is expected. Then ask Claude: "Use the Kirat extension and call kirat_status."
Claude Code: unzip the download and move the extracted kirat folder into your home folder, so that ~/kirat/claude-code exists. Then, in a session, run /plugin marketplace add ~/kirat/claude-code and /plugin install kirat@kirat, and enter your keys when asked. The plugin needs Node.js 20 or later.
Claude Code cloud sessions: a one-time setup of your cloud environment. Paste the block from the download (cloud/setup.sh) at the top of the environment's setup script and add both keys as environment variables; we recommend the names KIRAT_OPENAI_API_KEY and KIRAT_GEMINI_API_KEY, which don't collide with a project's own OPENAI_API_KEY (Kirat also accepts OPENAI_API_KEY and GEMINI_API_KEY). If network access is Full, there is nothing to change; otherwise set it to Custom (in newer apps: Limited) with the package-manager default list, plus api.openai.com and the Kirat download host. It has limits; read them first: where it runs.
The slow part is creating the two provider API keys — their signup and billing forms.
Is this worth it for a one-person project?
If you make decisions you'd want to justify to yourself later, probably. If you make a handful of decisions a month, no: two browser tabs are free and serve you just as well, and we'd rather say so than take your money.
What if I disagree with the council?
Then you overrule it. It gives you a result; what you build is your call. The run above is that case: the majority said hybrid, and the plan went another way.
Can it replace having a colleague?
No. A colleague remembers the history of the project, spots the thing you didn't ask about, and is on the hook with you. None of that is on offer here.
What you get instead: for a question you ask, two other model families take positions independently, Claude adds its own, and the disagreement survives into the result. That covers one specific failure of working alone — never hearing a considered "no" — and leaves the rest of it untouched.
Can I use it to sanity-check an estimate?
As a challenge, yes. As a number, no. The models don't know your codebase or your client, and a confident figure from three of them is still a guess.
Where it earns its keep is the risk you didn't price in — the migration nobody mentioned, the integration that always takes longer than planned. Ask what could make the estimate wrong, not what the estimate is.
Can it run up a bill?
The budget guard reserves the worst-case cost before every call and blocks calls that could exceed your daily budget — $5 per UTC day by default, adjustable. It is best-effort, so also set spending limits at the providers. A separate provider project with its own limit is a good idea.
I do client work — what about NDAs and client data?
Kirat has no server and no telemetry; we never see your keys or your data. Everything you give it 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. Compared with using Claude alone, that adds two recipients — so minimize or redact client data first, and check that your NDA allows OpenAI and Google.
The longer answer, including licensing for client work, is on the agencies page.
Do I need to be a developer?
No. If you can install a Claude extension and create two API keys, you can use Kirat. Claude Desktop ships its own Node.js, so Desktop users install nothing else.
If you do build software with an agent: Kirat itself is 11,351 lines of TypeScript, source and tests, in 38 commits, and every commit was written by an AI agent (Claude), none by hand.
Which API keys do I need?
One OpenAI API key and one Google Gemini API key, both with billing enabled. You pay OpenAI and Google directly for what you use. Claude is the third voter — it submits its own position — so there is no third key.
Kirat also runs with one key. Then the council has one external member, and every result names the missing key.
You enter the keys in Claude's own install or settings dialog. Claude Desktop stores them in the macOS Keychain or the Windows Credential Manager. Claude Code uses the macOS Keychain, and on Windows and Linux a credentials file readable only by your user. In Claude Code cloud sessions, the keys are environment variables of your cloud environment, readable by everyone who uses it.
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.