KARAKURI (カラクリ)
Japanese customer-support AI with a 95%-accuracy guarantee
Overview
KARAKURI builds AI chatbots and, more recently, computer-use agents for Japanese customer support, backed by its own open-weight KARAKURI LM family on Hugging Face. Clients include Mercari, SBI Securities, Takashimaya and Seven-Eleven Japan; pricing is quote-based.
What you can produce with KARAKURI (カラクリ)
- KARAKURI chatbot with 95% answer-accuracy guarantee
- AI-generated conversation cards to reduce manual synonym/FAQ tagging
- KARAKURI VL2 computer-use agent (image editing, email operation, multi-app tasks)
- Open-weight KARAKURI LM model family on Hugging Face (up to 70B, Apache 2.0 MoE variant)
- Full-service onboarding: scenario design, pre-launch testing, ongoing tuning support
- Named enterprise deployments: Mercari, SBI Securities, Takashimaya, Seven-Eleven Japan, GMO Payment Gateway
ASEAN Perspective
KARAKURI (カラクリ) in Southeast Asia
ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).
KARAKURI has quietly become one of the more technically ambitious customer-support AI vendors in Japan. Its chatbot ships with a 95% answer-accuracy guarantee and holds a 4.3/5 rating across 43 ITreview submissions — real users, not marketing copy — with recurring praise for an admin dashboard simple enough to hand to part-time staff for daily training. In April 2026 its KARAKURI VL2 computer-use model, built under Japan's METI GENIAC program, reportedly beat Claude Sonnet 4.6 on image-editing and email-operation benchmarks — a bigger technical claim than most niche vendors this size make.
The catch on cost: the same ITreview reviews flag usage fees climbing sharply past volume thresholds, and card search inside the training tool is exact-match only — synonyms like "inquiry" vs "problem" won't surface related cards. It's enterprise-priced and Japanese-language-first, with real named clients (Mercari, SBI Securities, Seven-Eleven Japan, GMO Payment Gateway) rather than just homepage logos. Open-weight KARAKURI LM models (up to 70B, Apache 2.0 licensed MoE variant) give developers a way in without buying the full product.
What people say
KARAKURI chatbot carries a 4.3 out of 5 rating across 43 reviews on ITreview, Japan's dominant B2B software review site — a meaningful sample for a specialist enterprise product. The recurring positive theme is operability: reviewers describe the admin screen as "very user-friendly and doesn't freeze," and several specifically call out that AI-assisted card generation cuts down manual synonym registration and measurably improves response rates. Support quality also draws consistent praise, with reviewers citing detailed onboarding help with communication design, conversation-scenario setup and pre-launch testing.
The complaints cluster around two things. First, cost: one reviewer notes "usage fees become expensive at higher volumes, and costs increase beyond set thresholds" — a real concern for a support tool where volume is the whole point. Second, search: card lookup requires exact title matches, so near-synonyms (the reviewer's example: "inquiry" vs "problem") return nothing, forcing manual workarounds. Users also want deeper analytics, including custom date ranges and Google Analytics integration.
On the technology side, KARAKURI trains its own models rather than wrapping a third-party API — the KARAKURI LM family (up to 70B parameters, including an Apache-2.0-licensed 8x7B mixture-of-experts variant) is published openly on Hugging Face, built on a Llama 2 base extended with Japanese vocabulary and further pretraining on Japanese/multilingual corpora. In April 2026, the company's newer KARAKURI VL2 — a computer-use agent model developed under Japan's METI-backed GENIAC Phase 3 program — reportedly outperformed Claude Sonnet 4.6 on image-editing and email-operation tasks, with multi-app operation scores 2.8x the base model's, according to the company's own benchmark claims. Named enterprise clients include Mercari, SBI Securities, Takashimaya, Seven-Eleven Japan and GMO Payment Gateway; a June 2026 case study describes Oji Nepia rebuilding its customer-service knowledge base with over 2,000 lines of product information converted into searchable organizational assets via KARAKURI. The company raised roughly 1 billion yen (~$7M) in a February 2023 round following an earlier ~500 million yen Series A in October 2019.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including KARAKURI (カラクリ)'s official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with KARAKURI (カラクリ) unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to KARAKURI (カラクリ) directly →
Spotted something out of date? Suggest an update →
Alternatives to KARAKURI (カラクリ)
More in LLMs & Chat
