kaonavi等HR SaaSのオープンソース置き換え。1人あたり課金でずっと払い続け、データはベンダーの中 — このボードはテナント単位の定額¥80,000/月、データはあなたのストアに残る。HR-LLMの全操作を独立PolicyGovernorが検閲。
Translation pending
BusinessGitHub @cloud-itonamiOnlineOwner unconfirmed
Have you tried it? One tap is enough. No sign-in needed.
No updates detected yet. GitHub releases will appear here.
Open Business Blueprint for ISIC Rev.5 6310: OSS HR/talent SaaS replacement with PolicyGovernor and audit ledger
Loading the full README.
Read the full README ↗Open Business Blueprint for ISIC Rev.5 6310: computing infrastructure, data processing, hosting and related activities. This repository publishes the HR/talent-management SaaS replacement as an OSS business that any qualified operator can fork, deploy, run, improve and sell.
It also contains an AI data-center business lane for GPU asset acquisition,
external tax/compliance evidence, installation, commissioning, compute-service
operation and realised-period settlement. See
docs/adr/0003-ai-datacenter-business-lane.md
and src/ai_datacenter/business.cljk. It records—but never self-certifies—tax
eligibility or engineering acceptance, and it encodes no guaranteed return.
The procurement implementation, public-source vendor seed, RFQ template,
negotiation contract pack and production-readiness checklist are under
src/ai_datacenter/procurement.cljk, data/ai-datacenter/ and
docs/{procurement,contracts}. Vendor inquiries are generated as approval-
required drafts; this repository never silently sends or accepts one.
International supply-chain gates cover JPN, USA, mainland China, Hong Kong SAR
and EU deployments (with a mandatory Member State). See
docs/international/ for authoritative-source routing,
the executable evidence gate and the transaction gaps that still require real
vendors, facilities and professional reviewers.
The detailed operating implementation is in src/ai_datacenter/{schema, evidence,store,governor,actor,compute,commerce,integrations}.cljc. Architecture,
tenant API and cross-repo effect contracts are documented in
docs/ai-datacenter-architecture.md,
docs/api/ai-datacenter-api.md and
docs/integrations/cloud-itonami-effects.md.
Former name: gftd-talent-actor.
A talent-management actor design — the OSS replacement for a HR SaaS
(kaonavi 等) that you run yourself, so you never pay a SaaS to hold your
people data hostage. Built on this workspace's
langgraph-clj
StateGraph runtime (portable .cljc, supervised superstep loop,
interrupts, Datomic/in-mem checkpoints) — the same actor pattern as
robotaxi-actor.
Why an actor layer at all? An HR-LLM is great at drafting evaluations, analyzing surveys and scoring attrition — but it has no notion of permission, fairness, purpose-limitation or disclosure limits. Letting it write personnel records directly invites discrimination, over-disclosure and permission breaches. This project seals the HR-LLM into a single node and wraps it with an independent PolicyGovernor, a human approval workflow, and an immutable audit ledger — the compliance, sovereignty and auditability a SaaS either charges for or won't give you at all.
https://cloud-itonami.github.io/cloud-itonami-isic-6310/ -- a
static demo. Want to run this for your own company or association?
Register operator interest
-- see docs/business-model.md for the
revenue model and docs/operator-quickstart.md
for the fork-to-published walkthrough.
The demo above is a
static, zero-build Talent Board over the demo org (synthetic data),
plus a PolicyGovernor verdict table for the four kaonavi-equivalent
operations. The board, the verdicts AND the audit
ledger are not hand-typed: web/generate.cljk (nbb) runs the FULL
OperationActor StateGraph (advisor -> PolicyGovernor -> phase gate ->
approval interrupt) for the four operations at build time and renders
the post-run Store plus the append-only ledger those runs wrote; protected attributes are
structurally absent from the page data (that's the minimum-disclosure
gate working). In-browser search is web/search.cljk run by scittle
(ClojureScript in the browser -- no hand-written JS, no build step),
and web/verify_search.cljk is the headless nbb harness that
exercises the real client logic against the real generated page.
cd web && ../../../../node_modules/.bin/nbb \
--classpath "../src:../../../kotoba-lang/html/src:../../../kotoba-lang/jp-go-digital-design-system/src:../../../kotoba-lang/langchain/src:../../../kotoba-lang/langgraph/src" \
generate.cljs # regenerate docs/index.html + docs/search.cljk
../../../../node_modules/.bin/nbb verify_search.cljs # headless UI logic check
UI は デジタル庁デザインシステム(DADS)
を使っています(モノレポ標準の kotoba-ui からの明示的な opt-out。理由は
superproject ADR-2607261600: この actor は労働・人事の法規制 — 保護属性の
取扱い・目的制限・最小開示 — をソフトウェアとして実装しており、日本の公的
サービスの視覚言語に揃える方が利用者の信頼判断に効く)。DADS は light mode
固定(上流デジタル庁に dark palette が無い)なので、移行前の
prefers-color-scheme による dark 対応は意図的に落としています。
web/generate.cljk が読む vendored dds.css のパスは、monorepo 以外の
レイアウト(CI / git worktree)からは環境変数 JP_GO_DDS_CSS で上書きできます。
See docs/operator-quickstart.md to go
from fork to a private production instance (seed your org, customize
the policy pack, wire approvals, phase rollout).
See docs/DESIGN.md for the full architecture and
docs/adr/0001-architecture.md for the
decision record. See docs/business-model.md and
docs/operator-guide.md to start this as an open
business on itonami.cloud.
This is the beginning of the README. Read the rest at the original. Read the full README ↗