# kwant

> Kwant is an AI agent workspace where bots only report tasks as complete after running their own verifications. It is built for developers and teams who need cited web research, payment confirmations, and auditable results from autonomous agents.

- Type: AI tool
- Page: https://completeaitraining.com/ai-tools/kwant/
- Published: 2026-10-30
- Updated: 2026-11-04
- Author: Jeroen Erne
- Jobs: IT and development, Management, Finance
- Topics: Productivity, Open Source, Developer Tools
- Reading time: 2 min

### About kwant

kwant is an open-source AI team that runs on your own Mac, Linux machine, or VPS. It lets you assign work to bots drawn from 27 roles, and each message routes to the bot that fits the task. The core rule is that no task is marked complete until kwant itself verifies the work-bots split tasks into cards, hand them off, and test each other's output.

### Review

kwant addresses a specific failure mode in AI agents: the confident "done" that doesn't hold up. The tool doesn't take a bot's word for it. Instead, it runs its own verification step before a task card can turn Verified, which changes how you relate to the output. This review walks through what's actually in the product right now, what's planned, and where the friction points sit.

#### Key Features

- **Verified-by-kwant checks.** A task card only flips to Verified when kwant runs the check itself. Bots can't self-report completion.
- **Cited answers with exact-quote links.** Web-sourced facts come with a link that opens at the specific quote. Missing or lookalike quotes get refused.
- **Truth Ledger.** When two sources disagree, kwant surfaces both versions and lets you pick.
- **Per-bot rules and site restrictions.** You specify what each bot should and shouldn't handle, and which websites its browser may open.
- **Flight School.** Bots earn more autonomy through scorecards, but only you can promote them. Payments always ask for confirmation-in the shell and in the bot's own browser.

#### Pricing and Value

kwant itself is free and open-source under the AGPL license. It runs on the Claude, Codex, or Grok subscription you already pay for-kwant doesn't sell its own AI credits. The cost view in Insights shows what each verified result actually cost, and it displays "Not reported" rather than a fake $0 when cost data isn't available.

#### Pros

- Verification is built into the workflow, not bolted on as a chatbot's self-report.
- Quote-level citations make it straightforward to check a bot's research without re-reading the whole source.
- Runs locally or on your own VPS, so data stays on your machine.
- Payment prompts appear even behind a plain "Continue," reducing surprise API charges.
- Open-source code means you can inspect how checks and routing actually work.

#### Cons

- Bots can still miss context that wasn't in the material they opened-like unread thread replies-and kwant's quote check alone won't catch that.
- Requires you to bring your own API subscription; there's no bundled AI access for people who don't already pay for one of the supported models.
- Not well suited for users who want a single-agent chat experience. The multi-bot, card-based workflow adds overhead that a simple Q&A tool avoids.

kwant fits teams and solo operators who've been burned by AI agents that claim completion without evidence. It's built for work where the cost of a false "done" is high-QA testing, research with citable sources, or multi-step tasks that need handoffs. If you don't already pay for Claude, Codex, or Grok, the tool won't do much for you, and the verification model assumes you're willing to review disagreements rather than receive a single smoothed-over answer.

---

This is the Markdown form of https://completeaitraining.com/ai-tools/kwant/. The whole site is mapped at https://completeaitraining.com/llms.txt.