Kentaro

Kentaro is the open-source, read-only Agentic Maysano Harness. It gives you a working local portfolio briefing and sourced dashboard immediately, then lets you opt in to a bounded read-only Maysano source without turning Kentaro into a system that acts on your behalf.

Learn more at maysano.com/kentaro.

Install

Kentaro requires Python 3.11 or later.

python -m pip install kentaro

To update an existing installation, run python -m pip install --upgrade kentaro.

Start in Codex

After installing, create an empty local project folder in Codex. In its first chat, send this message exactly; Kentaro handles the technical setup on the learner's behalf:

Set up Kentaro in this project. Run kentaro setup codex, then kentaro doctor, on my behalf. Do not clone the Kentaro repository. When the checks pass, give me a short plain-language summary with Setup, Health check, and What happens next. Confirm the 13 skills, local workspace, ODPS validation, and visual canvas; then tell me to start a fresh Codex conversation here before we begin the DesignPortfolio, with one ready-to-send business prompt.

Codex runs the installed kentaro setup codex and kentaro doctor commands in that folder. It installs the 13 skills in .agents/skills/, adds Kentaro's bounded AGENTS.md guidance, and creates the local workspace while preserving existing user instructions and changed skills.

When Codex reports that the health check passed, start a new conversation in the same folder. Codex loads AGENTS.md and skills when a conversation starts, so the setup conversation cannot reliably use the instructions it just installed. In the new conversation, start with a short business prompt, for example:

Our bank's customer risk assessments take too long. Help me improve this process.

Kentaro creates a local, evidence-labelled DesignPortfolio from that business input, then guides the learner through the next highest-priority design question. Learners do not clone the Kentaro repository, install skills, or type Kentaro commands themselves.

Run the demo

Run the complete synthetic demonstration with no account, configuration file, repository checkout, API key, or network request:

kentaro demo

It prints a sourced portfolio briefing and writes a dashboard to a new timestamped file under .kentaro/workspace/. Kentaro creates that directory automatically and prints the complete local path.

Use Kentaro locally

The normal local commands use the packaged synthetic portfolio context:

kentaro portfolio brief
kentaro dashboard

kentaro dashboard always creates a fresh timestamped HTML file in .kentaro/workspace/; it never overwrites an existing dashboard.

Design a local business and product portfolio

Kentaro v0.2 provides a deterministic local DesignPortfolio for objectives, use cases, AI products, data products, dependencies, evidence claims, readiness gaps, and reports. It is not an authoritative Maysano portfolio and never contacts a model or service in the standard path.

kentaro design init --title "Permit Service Transformation" --portfolio-id permit-service --output .kentaro/workspace/design/permit-service.json
kentaro design add-object --input .kentaro/workspace/design/permit-service.json --id objective-speed --kind objective --name "Faster permit decisions" --purpose "Reduce avoidable delay" --owner "Service director" --lifecycle-status proposed
kentaro design validate --input .kentaro/workspace/design/permit-service.json
kentaro design analyze --input .kentaro/workspace/design/permit-service.json
kentaro design report --input .kentaro/workspace/design/permit-service.json --output-dir .kentaro/workspace/design/permit-service-outputs

Generic commands never select a case automatically. The optional --scenario udemy-masterclass starter and the legacy masterclass commands explicitly select the course package.

Run the fictional Masterclass scenario

The Integrated Data & AI Product Management Masterclass is a fictional scenario on the same standard Kentaro installation. It uses a Customer Risk Intelligence case, keeps AI products distinct from data products, and does not contact Maysano, Ollama, or a hosted model.

python -m pip install kentaro
kentaro masterclass init
kentaro masterclass validate --input .kentaro/workspace/masterclass/customer-risk-intelligence/portfolio.json
kentaro masterclass report --input .kentaro/workspace/masterclass/customer-risk-intelligence/portfolio.json --output-dir .kentaro/workspace/masterclass/customer-risk-intelligence/outputs
kentaro masterclass validate-odps --artifact .kentaro/workspace/masterclass/customer-risk-intelligence/customer-profile.odps.yaml

The reports and canvas label proposed, assumed, and unknown information. SDK validation does not establish production readiness, business approval, compliance, or data availability.

Connect a read-only Maysano source

Live access is explicit and opt-in. Configure a Maysano-issued read-only key through the guided local setup, then select the source for an individual read:

kentaro configure maysano
kentaro portfolio brief --source maysano

The key is stored only in Kentaro's ignored local workspace and is never printed. The pilot source is GET-only. It does not currently provide owners, decisions, risks, evidence, approval history, or milestones; Kentaro labels those fields as unavailable rather than inventing them.

Safety boundary

Kentaro v0.1 and v0.2 retrieve bounded sourced context, validate local versioned governance and learning objects, generate local drafts and reports, classify proposed requests, and export local evidence. They do not create external records, write to Maysano or a business-context graph, execute business actions, request or process approvals, or claim immutable or compliance-grade audit. The optional local draft commands create local artifacts only; a hosted provider requires explicit acknowledgement before it receives source content.

Maysano remains the system of record for portfolio state, approvals, graph updates, tenant controls, authoritative revocation, retention, and platform audit. An allowed classification means only that a request is structurally in scope; it is not permission to execute.

Advanced configuration and governance

TOML profiles are an advanced team configuration mechanism for named environments, custom API endpoints or workspace paths, model declarations, organisation policies, and non-default profiles. They are never needed for the normal local experience.

To create proposed local data-product, use-case, or objective drafts from Markdown or text, use the SDK included in the standard installation. These local artifacts align with the Open Data Products (ODPS) family of specifications:

kentaro data-product draft ./business-context.md
kentaro use-case draft ./business-context.md
kentaro objective draft ./business-context.md

Validate a saved local draft without generating anything:

kentaro data-product validate ./data-product.yaml
kentaro use-case validate ./use-case.yaml
kentaro objective validate ./objective.yaml

The default provider is local Ollama. A hosted provider requires an explicit provider configuration and --allow-hosted-content, because it may receive the source document. Each result remains a local draft; it never creates a Maysano record or grants authority.

Detailed material is kept out of onboarding:

• Getting started
• Advanced configuration
• Local draft artifacts
• Local business design
• Masterclass Companion
• Governance and delivery object model
• Architecture
• Security and data boundaries
• Developer guide

MIT License © 2026 Maysano.