Bubo Inc. Launches QA Tool 'Nioh' to Evaluate AI-Written Code Before Release; Limited Beta Recruitment Begins Ahead of Official Launch via MCP-Compatible AI Clients

In software development environments where AI generates both code and tests at scale, the most labor-intensive task today is evaluating whether AI-written code is ready for release. Despite passing tests, defects may be overlooked, requirements missed, or compatibility issues discovered only after deployment. While AI excels at volume, determining correctness still relies on human judgment.

Bubo Inc. (Headquarters: Chuo-ku, Osaka City, Osaka Prefecture; CEO: Hiroyuki Watanabe), a specialist in software testing and quality assurance, announced in July 2026 the launch of 'Nioh (Nioh / Pronunciation: Nioh),' a tool that externalizes its internal QA methodology for assessing correctness. Nioh is accessible via MCP (Model Context Protocol)-compatible AI clients such as Claude and Cursor. Under the concept of 'Bringing Quality Intelligence to the Gate of Development,' Nioh covers requirement reviews, test design, validation of AI-generated code, and root cause analysis in operations. These tools are directly derived from Bubo’s own daily development practices.

The official version will be offered as a paid subscription. Ahead of that, Bubo is now accepting a limited number of companies for an early beta program on an interview basis.

Background: AI Can Generate Volume, But Humans Still Judge Correctness

It's often said that 'AI will take our jobs.' However, in 2026 development environments, the opposite is happening: the time required for humans to manually verify AI-generated code and tests is increasing.

Bubo considers this 'work of judging correctness' to be the core of future QA. The company has packaged its own internal QA procedures into tools accessible from local AI clients. In the beta phase, Bubo will collaborate with participating companies to evaluate how effectively these tools integrate into their workflows.

What It Can Do: End-to-End Support from Upstream to Operations

Nioh provides tools across all stages of software development. The overall structure follows the V-model development process.

Requirements Definition: Identify missing, ambiguous, or undefined requirements before implementation (e.g., requirement review, requirement formatting).

Design: Detect gaps in impact analysis and exception handling, and define necessary test perspectives and priorities during design (e.g., design review, test design, test prioritization, traceability checks).

Implementation & Code Review: Verify that AI-written changes meet requirements and do not break existing functionality, catching vulnerabilities to unexpected inputs and API incompatibilities during implementation (e.g., AI change validation, robustness checks, API compatibility checks, completion criteria checks).

Testing: Identify tests that pass but miss defects, or tests that are fragile and unstable (e.g., test effectiveness scoring, test code quality checks, flaky test detection).

Release & Operations: Conduct root cause analysis on reported bugs and feed insights back into preventive measures and additional test cases (e.g., defect analysis).

For example, asking the AI change validation tool 'Does this PR meet requirements?' will result in a comparison of changes, requirements, and tests, highlighting missing requirements and unnecessary modifications.

Two Usage Methods: Call in Conversation or Automate as a CI Gate

Interactive use: During development, call the tool from your AI client (e.g., Claude, Cursor) and receive real-time reviews and feedback.

Automated execution in CI: Tools that return pass/fail results can be integrated into CI systems like GitHub Actions to serve as automated 'quality gates' that block non-compliant changes.

How Is It Different from Simply Asking a General AI?

You might think, 'Why not just ask an AI to review?' Nioh differs from general-purpose LLMs in three key ways:

Structured viewpoints: The tool embeds QA perspectives—what needs to be checked. Requirements are reviewed using USDM; test design uses techniques like boundary value analysis and decision tables. Results are consistent and not dependent on prompt quality.

Deterministic and reproducible: Many tools use static analysis, ensuring identical outputs for identical inputs. This avoids the common LLM issues of inconsistent responses and hallucinations, reserving AI calls only for necessary cases.

Pass/fail output enables gating: Results are returned in structured formats (e.g., findings, scores, pass/fail), enabling integration into CI pipelines to automatically block changes. Unstructured conversational AI output cannot be used this way.

In short, while general AI 'answers when asked,' Nioh is a tool that 'knows what and how to verify.' It encapsulates Bubo’s accumulated QA methodologies from real-world development and QA practice.

Availability and Early Beta Program

Nioh will be officially offered as a paid subscription. The current early beta program allows select companies to test the functionality ahead of the official launch.

Official Version: Planned as a paid subscription (pricing and model to be announced).

Early Beta: Limited, interview-based slots. Due to the interview process, the number of participating companies is capped. During interviews, Bubo will jointly assess which tools fit your workflow and the optimal rollout sequence, and provide setup and integration guidance. Beta usage is free. The beta will end upon official release, and participating companies will receive priority access to the official version.

Why Offer the Beta for Free?

Before setting a price for the official version, Bubo wants to evaluate how effective each tool is in real development environments and refine the product based on user feedback. For Bubo, this period is about validating real-world utility before pricing. Therefore, they are inviting companies willing to provide candid feedback through a free, limited program.

How Is Confidential Code Protected? (BYOK)

Bubo’s servers do not store any code, specifications, or data passed to the tools. Bubo will not use this data for training.

LLMs are accessed using the user’s own API keys (BYOK—Bring Your Own Key), calling providers chosen by the user. Data handling terms are governed by the user’s agreement with the provider. Currently, tools using LLM keys support Claude and Gemini, with other providers to be added gradually.

Since LLM tools use BYOK, API usage fees from the LLM provider are incurred separately under the user’s provider contract (in addition to Bubo’s service fee). This applies to both beta and official versions. Tools that operate purely via static analysis incur no such API fees.

Tools are categorized into three types: (1) Tools that complete entirely via static analysis or aggregation without calling external LLMs (e.g., test effectiveness, robustness, resource leak, API compatibility, traceability); (2) Tools primarily using static analysis but optionally combining with LLMs; (3) Tools using LLMs for judgment or generation (e.g., requirement review, test design, AI change validation). Users can control the scope of data sent to AI.

Which tools send data to external LLMs and which do not, communication