Open-source AI team for your Mac, Linux or VPS. Hire bots from 27 roles; each message goes to the bot that fits, and you see why. Bots split work into cards, hand off and test each other, and a task turns Verified only when kwant checks it. Web facts come with links that open at the exact quote. Bots earn autonomy in Flight School, but payments always ask, and you choose which sites each bot may open. Runs on the Claude, Codex or Grok you already pay for.
AI agents are great at saying "done". They're worse at being done. I kept getting confident reports about tests that never ran and links that said something else. So kwant is built around one rule: nothing counts until it's checked.
What you get:
🗂️ Mission Control: bots split the work into cards, hand off and test each other
✅ Verified means verified: a card turns Verified only when kwant runs the check itself
🔗 Cited answers: web facts come with a link that opens at the exact quote. Missing or lookalike quotes are refused
⚖️ Truth Ledger: when two sources disagree, you see both and pick
🎓 Flight School: bots earn more autonomy with scorecards, and only you promote them
💳 Payments always ask: in the shell and in the bot's own browser, even behind a plain "Continue"
🛡️ Rules per bot: say what each bot should and shouldn't take, and which sites its browser may open
📊 Insights: what each verified result actually cost ("Not reported", never a fake $0)
🧠 Who answers: each message goes to the bot that fits, and you see why. Wrong bot? Pick another one, and kwant remembers. Laya picks the model and effort for bots on auto, so quick asks stay cheap
✨ Suggested routines: kwant notices your habits and offers to automate them
🏠 Spaces: work and home stay apart
It's one signed app for Mac and (new this week) Linux, or a VPS. It runs on the Claude, Codex or Grok subscription you already have. Your data stays on your machine, and it's open source (AGPL).
'Great at saying done, worse at being done' is exactly it — the failure I keep hitting is a confident summary of tests that never actually ran. The rule that a card only turns Verified when kwant runs the check itself, instead of trusting the bot's own report, is the whole product for me. Refusing missing or lookalike quotes is the same instinct applied to research, and I appreciate 'Not reported' instead of a fake $0 in the cost view — most tools would just print the zero.
Worst one this week: the morning summary told me my cofounder hadn't answered someone on Slack, and he had, in the thread replies it never opened. Nothing it read was wrong, it just didn't read enough. Would kwant's check catch that, or only a source that says something else?
About kwant on Product Hunt
“AI bots that check each other's work before saying "done"”
kwant launched on Product Hunt on November 3rd, 2026 and earned 77 upvotes and 4 comments, placing #19 on the daily leaderboard. Open-source AI team for your Mac, Linux or VPS. Hire bots from 27 roles; each message goes to the bot that fits, and you see why. Bots split work into cards, hand off and test each other, and a task turns Verified only when kwant checks it. Web facts come with links that open at the exact quote. Bots earn autonomy in Flight School, but payments always ask, and you choose which sites each bot may open. Runs on the Claude, Codex or Grok you already pay for.
kwant was featured in Productivity (664.7k followers), Open Source (69k followers), Artificial Intelligence (482.7k followers) and GitHub (41.5k followers) on Product Hunt. Together, these topics include over 361.1k products, making this a competitive space to launch in.
Who hunted kwant?
kwant was hunted by Oleks. A “hunter” on Product Hunt is the community member who submits a product to the platform — uploading the images, the link, and tagging the makers behind it. Hunters typically write the first comment explaining why a product is worth attention, and their followers are notified the moment they post. Around 79% of featured launches on Product Hunt are self-hunted by their makers, but a well-known hunter still acts as a signal of quality to the rest of the community. See the full all-time top hunters leaderboard to discover who is shaping the Product Hunt ecosystem.
Want to see how kwant stacked up against nearby launches in real time? Check out the live launch dashboard for upvote speed charts, proximity comparisons, and more analytics.
Oleks
Hi PH 👋 I'm Oleks, maker of kwant.
AI agents are great at saying "done". They're worse at being done. I kept getting confident reports about tests that never ran and links that said something else. So kwant is built around one rule: nothing counts until it's checked.
What you get:
🗂️ Mission Control: bots split the work into cards, hand off and test each other
✅ Verified means verified: a card turns Verified only when kwant runs the check itself
🔗 Cited answers: web facts come with a link that opens at the exact quote. Missing or lookalike quotes are refused
⚖️ Truth Ledger: when two sources disagree, you see both and pick
🎓 Flight School: bots earn more autonomy with scorecards, and only you promote them
💳 Payments always ask: in the shell and in the bot's own browser, even behind a plain "Continue"
🛡️ Rules per bot: say what each bot should and shouldn't take, and which sites its browser may open
📊 Insights: what each verified result actually cost ("Not reported", never a fake $0)
🧠 Who answers: each message goes to the bot that fits, and you see why. Wrong bot? Pick another one, and kwant remembers. Laya picks the model and effort for bots on auto, so quick asks stay cheap
✨ Suggested routines: kwant notices your habits and offers to automate them
🏠 Spaces: work and home stay apart
It's one signed app for Mac and (new this week) Linux, or a VPS. It runs on the Claude, Codex or Grok subscription you already have. Your data stays on your machine, and it's open source (AGPL).
What would you hand to a team like this first?