Techtouch

Named third party systems are reached by working on their screens, and the absence of integration is marketed as the feature. The AI Hub page's headline reads API不要 (no API required) alongside システム改修なし (no system modification), and the capability heading is 外資SaaSからスクラッチシステムまで、API不要・改修不要でシステムをAIネイティブ化. Where the agent changes something in another vendor's system, it types into the form.

The reach is real. The product works across third party systems a customer already runs, with ERP, expense settlement and indirect procurement each given a solution page. The site names CRM (Salesforce), ERP and HR systems (SAP, SuccessFactors), expense and procurement tools (Concur, Coupa) and content storage (Box).

One connection at the data level is documented on the AI side: 貯社RAGと業務システムを繋ぐことで (connecting the customer's own RAG to their business systems), so an AI chat can be called directly from within the system. AI OCR ingests uploaded documents, and distribution runs through the AWS Marketplace and the Coupa App Marketplace.

No connector catalog or webhook appears anywhere, and no authenticated API call into a third party system or named surface for calling tools either. The customer's business systems are canvases, not callees, and the RAG connection is described in one clause without a page of its own.

Sourcetechtouch.jp/ai_hub product page and FAQread 2026-09-12

The agents carry out real work of several steps inside a task, with a documented fallback to guided manual entry, but nothing chains them across systems. The vendor's framing is 一連の業務プロセスをAIが代行 (the AI performs a series of business process steps on the user's behalf), and the worked example bears it out. The expense agent takes an uploaded receipt, reads date and amount by AI OCR and applies the customer's input rules.

It makes a judgment where categories are mixed, such as separating lodging cost from lodging tax, computes the result and writes it into the form. The validation agents check an entry against the company's rule set, raise an alert and guide the user to whichever application is actually required. A fallback runs between the two products: when an irregular case arises, the AI agent performs the standard scenario and the DAP's guidance takes over for manual input on the exception.

Each agent works within one screen of one system. There is no flow builder, branching or routing engine, and no approval step inside a process Techtouch owns, long running state or sequence across systems, which follows from a browser extension with no API. The agents are listed as independent units by domain rather than as stages that call each other, and the participants are the employee and one agent. Where a true end to end workflow exists, it belongs to the customer's underlying system.

Sourcetechtouch.jp/ai_hub agent descriptions and DAP fallback sectionread 2026-09-12

The agents connect to the customer's own retrieval system and company rules, but Techtouch keeps no corpus of its own. The AI Hub's first capability lists 生成AI、AIOCR、RAGなど最新機能をシステムに後から埋め込み (generative AI, AI OCR and RAG embedded into the system after the fact).

A path into the customer's own corpus is documented: 貯社RAGと業務システムを繋ぐことで、あらゆる業務領域でシステムから直接呼び出せるAIチャットを実現 (connecting your company's RAG to your business systems so an AI chat can be called directly from inside the system), and the FAQ confirms an existing internal FAQ chat can appear on the system screen.

The company rule set is a second standing corpus: the validation agents check entries 自社の労務ルールに (against the company's own labor rules), and the procurement agent judges operating rules too complex to encode in the system itself. Those rule sets persist between runs and are queried.

The maintained retrieval structure belongs to the customer: the vendor's word is 貯社RAG, your company's RAG, which Techtouch connects to. No ingestion pipeline, index or curation surface of Techtouch's own is documented, and no article lifecycle or citation behavior either. Screen context is read live, for each request. The RAG connection is stated in a single clause, and the rule sets are described through the agents that use them rather than as a knowledge product.

Sourcetechtouch.jp/ai_hub capability list, agent descriptions and FAQread 2026-09-12

Irregular cases go to a person, but no review of what the agent writes before submission is documented. Techtouch writes イレギュラー時には人が介在して補う設計で、AIによる間違いを防ぎつつ業務を遂行します (a design in which a person intervenes on irregular cases, preventing AI mistakes while the work still completes), and states the principle as "Human in the loopでも精度を補完し、必ず完結させる".

For routine work the agent performs the standard scenario, and when an irregular case arises the DAP guidance takes over so the employee completes the step by hand. The person picks up those cases with the guidance tooling in front of them. Administrators also set usage rights and ceilings per role and position from the management console, which bounds what the agents may do and for whom.

Nothing says the agent's entries are reviewed before submission. The product is sold as 業務代行, doing the work on the employee's behalf, with the AI operating the screen directly to enter what it generated, and no review queue, approval step or confirmation prompt appears, nor a diff view or a setting for how much autonomy the agent has.

The intervention that is documented starts when the agent itself detects an irregular case, which routes exceptions rather than giving a person a gate to hold. The validation agents for attendance and quotation applications work the other way round: the person acts and the AI checks the entry.

Sourcetechtouch.jp/ai_hubread 2026-09-12

ISO certifications that anyone can verify sit alongside a named set of access controls for customers. Under a heading reading 認証, the page displays ISO/IEC 27001 and ISO/IEC 27017, each linked to the certification body's public register for the certificate's scope; ISO 27017 is the security control standard for cloud services. Service operation runs at AWS Foundational Technical Review accreditation quality against a published service level objective.

The access controls are a named feature set, 安心してサービスを利用するための機能, listing SAML SSOを用いた認証 (SAML SSO authentication), IPアドレスによるアクセス制御 (IP address access control) and ロールベースの権限管理機能 (role based permission management), plus operation backup. The AI Hub adds its own admin layer: member and permission management sets usage rights and ceilings by role and position, and SSO connects to the company's directory service.

The security program is specific. External security vendors and tools check the application and cloud configuration for vulnerabilities, Cloud Security Posture Management runs alongside, and third parties run periodic vulnerability diagnostics. Data moves over TLS 1.2 or above and is stored with algorithms recommended by Cryptrec, and authentication is layered.

Staff get secure coding training and a written AI usage guideline, their endpoints carry MDM and antivirus, and accounts are reviewed periodically. A security checklist workbook is published for download so a buyer can evaluate the product before adopting it without a call. No SOC 2 is claimed.

Sourcetechtouch.jp/information-security, with techtouch.jp/ai_hub admin featuresread 2026-09-12

An audit log of account activity goes to administrators, alongside usage reporting on the agent layer. The audit log is a named feature, 監査ログ対応, described as アカウントアクティビティの詳細な証跡を管理者に提供します (detailed evidence of account activity provided to administrators), so that 利用された内容を正確に把握することができます (an administrator can know exactly what was used).

The agents read screens and type into forms for employees, so a trail of AI usage per account is a record of what the agent did, and 証跡 is the term for audit evidence, not a usage counter. Usage reporting, 利用状況のレポート, gives administrators dashboards on how users engage with the AI. Both sit under the AI Hub's own 管理・分析機能 heading, so they cover the agent layer.

The DAP's analytics, which show how systems are used and where users struggle, measure the customer's workforce against its own software rather than the agent. The audit log is described in two sentences on a product page, with no published retention period, field list or export format, and no view of an agent's individual actions in a run.

Sourcetechtouch.jp/ai_hub administration and analytics featuresread 2026-09-12

What persists is configuration people author and usage analytics, not agent memory. Customers write guidance in a no code editor and it is shown over the target system; it persists because someone saved it, not because an agent kept anything.

The AI layer's rule sets work the same way: 自社ルールを学習したAI is an agent configured against the company's rules during a customization engagement of around two months, which is setup rather than accumulation.

Usage state is analytics on the customer's own operation, recording what users did, and resilience to UI change is an element matching property of the extension, not memory.

No named memory component appears, and no preference or correction per employee that the agents reapply, or decision trace carried between runs. There is no stated scope or lifetime, and no customer surface to inspect, edit, export or purge agent state. The design points away from memory: the agents read the current screen as context and act on it, so each run is bounded by the page in front of the user. Longer memory would live in the customer's own RAG, which Techtouch connects to rather than owns.

Sourcetechtouch.jp/ai_hub and techtouch.jp/information-securityread 2026-09-12

It runs as a cloud service with no named region, customer environment or residency option. The information security page describes service operation in four sections, reliability, vulnerability management, defense against unauthorized access and optional security features. It names AWS Foundational Technical Review accreditation quality, backup and restore, and encryption in transit and at rest, along with a published service level objective, but no region, data center location or residency choice. A marketplace listing is a way to buy, not a deployment option.

The absence is built in. The product's proposition is API不要・改修不要 (no API, no modification), with a browser extension and a page snippet as the only footprint, so there is nothing to host yourself, and no hybrid, single tenant or bring your own cloud option appears. The footer publishes 外国にある第三者への個人データの移転について, a notice required by Japanese privacy law that personal data is transferred to third parties abroad, which is the opposite of a guarantee to keep data local.

Techtouch sells to central government agencies, municipalities, independent administrative institutions and universities as a named segment, and Japanese public sector buyers routinely require domestic residency. Nothing published answers that requirement.

Sourcetechtouch.jp/information-security and techtouch.jp/ai_hubread 2026-09-12

A catalog of eight named agents covers four back office domains, each agent with a stated job. Accounting has 経費精算エージェント, which reads an uploaded receipt by AI OCR to extract date and amount, then fills the form to the customer's input rules and handles judgment cases such as splitting lodging cost from lodging tax, along with 請求書申請エージェント.

Attendance and labor has 勤怠申請バリデーションエージェント, which checks timesheet entries against the company's own labor rules and raises alerts, and 勤怠承認サポートエージェント. HR has 目標ブラッシュアップエージェント, which reviews an employee's objectives against SMART criteria, and 評価ブラッシュアップエージェント. Procurement has 見積申請バリデーションエージェント, which checks quotation requests in real time against operating rules too complex to encode in the system itself, and 請求書登録エージェント.

They are separate units, not stages of one workflow: an expense agent and an HR objective review agent share no pipeline, and a finance buyer can adopt the accounting pair without the HR pair.

They arrive prebuilt and proven, in Techtouch's framing 他社ですでに成果が出ているAIエージェントに貴社のルールを組み込ませて最適化 (agents already delivering results elsewhere, customized with your rules), drawing on operational knowledge from 300 large enterprises, with customization to each company's rules taking around two months.

The sales and core systems domain and the IT and information management domain both read Coming Soon, and WF申請バリデーション is listed as 提供予定 (planned).

Sourcetechtouch.jp/ai_hub agent catalogread 2026-09-12

An agent fires when an employee reaches the relevant screen, a real but narrow trigger on a single surface. The browser extension reads the page the employee is on, so an agent switches on in context: the expense agent when a receipt is uploaded, the attendance validation agent when a timesheet is filled in, the quotation agent when a procurement request is entered, each checked リアルタイムで (in real time). Being in a context is a genuine trigger, and the employee does not have to configure it.

Two things are missing. The trigger is always a person arriving somewhere: no event from a system change, schedule or webhook is documented, and no timer or inbound message either, and a browser extension runs only while a person has the page open.

And the only channel is the system screen: no Slack, Teams or email channel appears, nor a portal, phone or ticket channel, and the FAQ chat on offer sits on the system screen rather than in a messaging client. The DAP's guidance that switches on in context is the same mechanism, not a second one. The site never mentions scheduled or event driven runs, though no page rules them out.

Sourcetechtouch.jp/ai_hub agent descriptions and FAQread 2026-09-12

OpenAI models on Azure are the only ones named, and there is no choice. The AI Hub FAQ says 現在はAzure上のOpenAIモデルを利用できます (currently, OpenAI models on Azure can be used): a single provider on a single hosting platform, disclosed by the vendor, so the lack of choice is visible rather than inferred.

The same answer adds 今後、モデルの接続先を増強する予定です (there are plans to strengthen model connection destinations in future), a plan that confirms connections are singular today. It closes with ※ご希望のモデルがございましたら、お問い合わせよりお気軽にご相談ください (if you have a desired model, please consult us).

An invitation to raise it with sales is not a control the customer holds: there is no picker, admin setting or bring your own key path, no model assigned per agent, and nothing says such a request has been met. Nothing says Techtouch routes across more than one model internally either, and Azure is infrastructure, not a second model.

Sourcetechtouch.jp/ai_hub FAQread 2026-09-12

No developer surface is published: no API reference, developer portal or SDK appears in the navigation or footer, and no webhook documentation or MCP server either.

The footer is exhaustive, running from the media site, an engineering blog at tech.techtouch.jp and a separate product site for AI Central Voice to a support page, a status page and two PDF operational runbooks, alongside the terms, addenda and privacy notices, the information security policy and page, and a service level objective.

The login leads to an editor application, not a developer console, and tech.techtouch.jp is an engineering blog about how Techtouch is built, not a developer surface for customers. A no code editor is where customers author their own guidance, and marketplace listings are ways to buy; neither makes Techtouch callable from outside.

The design fits the absence. The product's stated proposition is API不要, that a customer needs no API to adopt it because the browser extension reaches the screen instead, so a vendor whose selling point is that integration is unnecessary has little reason to publish an inbound API. A help center sits behind the support page, and it may hold API documentation.

Sourcetechtouch.jp/ai_hub and techtouch.jp/information-security, navigation and footer read in fullread 2026-09-12

Reports show how the AI is used, but no way to test a change before it reaches employees is documented. The AI Hub ships 利用状況のレポート, dashboards that let administrators see how the AI is being used and act where uptake is thin, and the DAP layer's analytics show where users struggle inside a system. Techtouch markets the agents as 効果を実証済み (effect already demonstrated), drawing on operational knowledge from 300 large enterprises, though no result format is published. A live metric on use exists.

Nothing puts a change under test. No test environment, sandbox or dry run appears, nor a holdout, versioning of agent rules or a scored verdict from an evaluator. A customer's rules are built into an agent over around two months and then the agent runs, with no described run of a changed rule set against known cases and no before and after comparison.

Continuous improvement after release (リリース後も継続的な改善で使い勝手を磨き、納得いく成果を届けます), backed by domain professionals, is Techtouch's own people tuning the deployment inside an engagement, not a harness the buyer receives. The site is in Japanese with no documentation host, and the help center may describe a test capability.

Sourcetechtouch.jp/ai_hubread 2026-09-12

The AI reads the screen of systems Techtouch does not own and then operates that screen, through a browser extension. The AI Hub page lists as its first named capability 画面上の情報をコンテクストとしてAIが取り込み (the AI takes in the information on the screen as context), followed by AIが判断・生成した内容を、そのままAIが直接画面操作して入力 (the AI directly operates the screen to enter what it judged or generated). It reads the screen, decides and drives the screen, and the vendor states this as its own design.

Delivery is stated too, ブラウザの拡張機能を通して提供されます (the functions are delivered through a browser extension), which is why the underlying system needs no change. It runs inside systems Techtouch does not own, from external SaaS to systems built from scratch in house, and the worked example is an expense agent that reads an uploaded receipt by AI OCR and fills the form to the customer's input rules.

Changes to those interfaces are the central engineering risk for a product built this way. Techtouch markets its guidance as resilient to changes in the underlying interface and publishes a documented procedure for investigating and halting the browser extension's effects. The browser is the only modality here, and operating an interface it does not control is the whole design.

Sourcetechtouch.jp/ai_hub product page and FAQread 2026-09-12