What is a Kami?

A small AI that serves only one place or one group. The people there can inspect it, correct it and retire it at any time. It is a metaphor from Audrey Tang that borrows the shape of the Shinto idea of a "god of a particular place".

When people think about the future of AI, many picture one giant AI that knows everything in the world and runs everything. A Kami is an idea that aims for the opposite: placing many small AIs, each of which looks after just one place — a neighborhood association, a classroom, a small charity.

In short

A Kami is a small AI helper that serves only one place or one group. It answers to the people there; its records can be read, it can be corrected, and it can be switched off at any time [1][2].

It is an idea that Audrey Tang, Taiwan's former Minister of Digital Affairs, and colleagues have used since 2025 within the 6-Pack of Care, a framework for governing AI. It is the unit in which Civic AI (AI that is accountable to communities and organizations) is actually deployed.

For example

A Kami that runs on your own laptop. In June 2026, civic.ai published steps for starting a Kami. You run a language model on your own computer with the free app Ollama, and give OpenClaw, an AI agent (an AI that carries out tasks on its own), one sentence: "Please read civic.ai and become my local Kami." [1][3]

The Kami then first "writes three little notes about itself": roughly, what it promises, its name and manner, and whom it serves and which lines it must not cross. Then "it starts asking questions", such as "which place or group am I here for?", "What may I decide on my own, and what must stay with humans?" and "How do people correct me or switch me off when I get it wrong?" [1] The same page stresses: "ten minutes gives you a working Kami. The hard part comes after — earning your community's trust." [1]

A Kami that helped prepare a talk. In July 2026, at a Wikimedia gathering at the National Library of France (BnF) in Paris, Tang said the talk itself had been prepared with a Kami. As that Kami's "working conditions", she listed: "No source, no sentence"; "An agent that reads the commons owes the commons an edit" (the commons being knowledge shared by everyone); and, when edits collided, nothing was silently overwritten — "a human reconciled the conflict" [4].

A family Kami. In a June 2026 talk, Tang said that she and her younger brother had built a small system for the family's chat, "what we call a Kami". The goal, set by their mother, was that the father's conversations with it should "reduce his reliance on the screen and increase his peace of mind" [5].

How it works

Where the name comes from

Kami stands for "Knowledge Artefact Management Intelligence". At the same time, it deliberately echoes kami (神), the gods of Japanese Shinto: local deities that dwell in a particular river, forest or old tree, and whose existence is bound up with the health of that one thing. civic.ai writes: "The river's guardian has no ambition to manage the forest" [6][2].

The acronym, however, may have been added later. A record of a September 2025 talk mentions Kami as a "spirit of place", but no expansion of the acronym appears there. The earliest source in which the acronym can be confirmed dates from March 2026 [7][8].

Four properties

In the Paris talk, Tang described a Kami with four words [4]:

Bounded — "Trained for one purpose, not every purpose."
Local — its memory, source ledger and house rules "stay on a laptop with sixteen gigabytes of memory". The provider of the language model it uses "is replaceable".
Inspectable — "By the people it serves."
Retirable — it can be retired at any time. In Tang's words, "retirement is not poetry. It is a contract clause."

Why build them small

AIs might compete and expand, until in the end one AI controls everything. The sixth principle of the 6-Pack, "Symbiosis", tries to answer this fear of a "singleton" (a single ruler) [6].

The manifesto says the answer is not to have an even bigger AI take control. Instead, the AIs coordinate in two ways [6][2]:

Federation — "The AIs agree on a shared set of rules for how to interact peacefully"
Subsidiarity — "problems should always be solved at the most local level possible." For example, "The national-level AI shouldn't interfere with the city-level AI unless there's a problem the city truly cannot solve on its own."

The manifesto states clearly that being bounded is not a built-in property of AI. The Kami's "boundedness is not intrinsic but engineered: resource caps, sunset timers, non-expansion pacts, and fresh democratic authority for any scope change" [6][2]. It also says that "because people belong to many overlapping communities, no single Kami could dominate over the whole person" [6].

Not a hierarchy, but overlapping centers

When asked on the online forum LessWrong whether a society of Kami would be equal or hierarchical, Tang answered that it would be neither. A Kami dealing with climate at the scale of a continent and a Kami dealing with a local irrigation channel differ in size. But, Tang explained, that is a difference in the scope each looks after, not in rank; each Kami is strong only within its own scope [9].

Does it have a mind?

The civic.ai FAQ deliberately leaves open whether a Kami has a mind or consciousness. It then says that a Kami is infrastructure that people assemble and steer, and must not become a substitute for care between human beings [10].

What is debated

Is it right to borrow a religious idea? Tang herself limits what is borrowed. According to a report on one of her talks, Tang, who grew up in a Taoist family, explained that she is not Japanese and has not studied Shinto, and has clearly nothing to do with State Shinto, the imperial family or Yasukuni. What she borrows, she said, is only the sense of a small being tied to a specific place that people can come to know [11]. In the Paris talk, too, she said: "I borrow only the shape of that idea, with respect and no claim on anyone's religion" [4]. In a conversation with Hiroki Azuma, she contrasted the all-knowing, all-powerful God of monotheism with Kami as the "eight million gods" of Japan, which share the same written character (神) yet are something quite different [12]. Note that criticism of this borrowing by third parties is not yet found among the materials collected for this site.
"Local" does not mean "safe". civic.ai warns that a small model on your own machine can tell smooth, confident lies just as a large remote model can: "private does not mean honest" [1]. It recommends a test: ask about something that never happened, and check that it answers that it does not know [1].

Common misunderstandings

A Kami is not a particular software product. No independent implementation has been found. What civic.ai offers is a procedure for configuring existing AI agents [1].

Where it fits in Plurality

Kami does not appear in the book ⿻ Plurality (2024). (The editors of this site checked this by searching the Japanese, English and Mandarin editions of the book.) It is an idea brought in after the book came out, with the 6-Pack of Care (2025 onward), and it is Tang's claim, not the book's.

Still, there is a link to the book. At the start of chapter 2-2 the book quotes a poem Tang wrote in 2016, ending: "When we hear 'the singularity is near' — / let us remember: The Plurality is here." [13] Tang uses this line again and again when explaining Kami. In the manifesto she continues: "The singularity is a vertical vision. Plurality is a horizontal one. The future of AI is a decentralised network of smaller, open and locally verifiable systems — local Kamis, spirits of place." [6]

Articles to read alongside this one:

Civic AI — Kami is its unit of deployment
6-Pack of Care — Kami appears in the sixth principle, "Symbiosis"
Superintelligence — the "singleton" that Kami is set against
⿻ Plurality — the source of "the Plurality is here"

Sources

civic.ai, "Set up your own Kami" (2026-06-06) — https://civic.ai/kami/ ［CC0］
civic.ai Glossary — https://civic.ai/glossary/ ［CC0］
OpenClaw README — https://github.com/openclaw/openclaw ［MIT］
Audrey Tang, "Kami: Knowledge Artefact Management Intelligence" (BnF Paris, 2026-07-21) — https://audreyt.org/kami-mosaic.html ［CC0］
civic.ai, "Reimagining AI Alignment" (2026-06-02) — https://civic.ai/reimagining-ai-alignment/ ［CC0］
"6-Pack of Care: A Manifesto" — https://civic.ai/manifesto/ ［CC0］
LessWrong, "⿻ Plurality & 6pack.care" (talk notes from 2025-09-01) — https://www.lesswrong.com/posts/anoK4akwe8PKjtzkL/plurality-and-6pack-care
civic.ai, "Democracy Needs Civic AI" (2026-03-25) — https://civic.ai/democracy-needs-civic-ai/ ［CC0］
Comments on the same LessWrong post (2025-09-02) — https://www.lesswrong.com/posts/anoK4akwe8PKjtzkL/plurality-and-6pack-care
civic.ai FAQ, Q20 — https://civic.ai/faq/#faq-20 ［CC0］
Sarah Cao, "Civic AI and the Muscles of Care" (STAIR, 2026-06-27) — https://www.stairjournal.com/oped/2026/6/27/civic-ai-and-the-muscles-of-care-reflections-from-ambassador-audrey-tang
Conversation between Audrey Tang and Hiroki Azuma (2026-03-15, in Chinese) — https://archive.tw/2026-03-15-%E6%9D%B1%E6%B5%A9%E7%B4%80%E5%B0%8D%E8%AB%87 ［CC0］
Plurality, 2-2 "The Life of a Digital Democracy" — https://www.plurality.net/v/chapters/2-2/eng/ ［CC0］

We welcome error reports and questions ("What is…?"). Submissions become public issues.

Report an error in this article ・ Ask a question