Hoshin Kanri is the X-Matrix all your agents share: objectives, initiatives, KPIs and reviews, readable and writable through MCP. Each agent keeps its own access to the MES, QMS or ERP. When a deviation comes up, it follows the links to the objective it touches and proposes a point for review. The committee decides.
You. The milling shop reports the inspection camera is down for three weeks. What does it affect? Claude. The machine-vision quality initiative, which carries the customer non-conformity objective of 0.8% (1.4% today). Moving the target or compensating is your call. Shall I file it in the inbox, linked to that objective, for Monday’s review?
Hoshin does not plug into your systems. Each agent keeps its access and its rhythm; they all read the same X-Matrix, and what they file is for people to decide.
Your systems, your rights
Prepares reviews and summaries
Links quality issues to objectives
Watches indicators on its own schedule
Follows the plant matrix, under the group’s
Objectives, initiatives, indicators and the links between them, read through MCP. Score, chains and cascade are computed, not generated.
No line from here to your systems: the agents make that link.
Proposals, in the inbox
The agent then acts in its own tools, if you allow it.
Every leadership team knows the artifact: the X-Matrix built in Miro or Excel, where formatting took more time than the strategic thinking, frozen the day it was presented. Meanwhile the agents arriving in your company run on the opposite excess — piles of unstructured context: PDFs, exports, document dumps, chat history. Abundance of unstructured context is the enemy of agentic efficiency. Japanese manufacturers tackled this for human teams from the 1960s with policy deployment, and the X-Matrix later gave it one bounded, correlated page, from vision to daily work. That frame is exactly the structured context agents need: light enough to hold entirely, with explicit links, and kept alive by the people and agents who use it.
The X-Matrix: 3-5 year objectives (North), annual objectives (East), initiatives (South), KPIs (West) — linked by correlation strength in all four corners. Nothing enters the strategy without earning its place in the chain. Shareable by read-only link — and compact enough for an agent to hold entirely in context.
A 5-level meeting hierarchy from strategic reviews to daily stand-ups. Agendas, action items and decisions traced from vision to daily work — the catchball, structured. The review is where strategy deploys or dies.
5 computed metrics — coverage, correlation density, orphans, vision-to-KPI chains, leading/lagging indicator balance. One number that answers the only question that matters: would this strategy survive contact with reality?
Policy deployment aligns your 3-5 year vision with daily actions. Here is the loop — and what your agent does at each step.
3 to 5 breakthrough objectives for the next 3-5 years — the North of the X-Matrix. Few enough to mean something.
Your agentturns notes and ideas captured mid-conversation into candidate objectives in the idea inbox.
What must be true this year for the vision to stay credible (East). Negotiated with the teams — the catchball — not decreed.
Your agentchecks every annual objective actually supports a long-term one, and challenges those that support none.
Improvement initiatives (South) and the indicators that will prove they work (West) — an owner on every row.
Your agentflags orphan initiatives, unmeasured objectives and a lagging-heavy KPI set via the completeness score.
A 5-level meeting hierarchy, from strategic review to weekly sync. The review is where strategy deploys or dies.
Your agentprepares each agenda: pulls progress from your project tools, updates statuses, lists off-track KPIs and pending decisions.
Decisions traced, initiatives re-arbitrated, the matrix kept alive — hansei, the reflection that closes the loop.
Your agentdiagnoses how deeply each strategic theme is deployed and lists the arbitrations to make for the next cycle.
Hoshin Kanri ships an MCP server with OAuth 2.1 and API tokens. Connect an assistant in two minutes, or have an admin create a token for an autonomous agent on its own schedule. The agent reaches your MES, QMS, ERP or Jira through their MCP server or a bridge: Hoshin does not connect to them. Every teammate brings their own, your consultant too: same matrix, each with their own rights. The agent reads objectives, checks initiatives, tracks KPIs, prepares reviews and flags drift. It proposes; people decide. One simple rule:
Strategy, reviews and KPIs → Hoshin Kanri. Work orders, non-conformities, tasks and tickets → your own tools.
18 MCP tools · 9 guided prompts · idea inbox, deployment diagnosis, boardroom artifacts, public share links. Enterprise: federate it behind your company's Cortex Gateway as the steering backend, next to your business apps.
See it live — Mecanor, a fictional manufacturer with a fully populated X-Matrix: Open the live demo — or query it from your own agent (read-only MCP token in llms.txt)
Give this address to your own AI assistant and ask for a verdict:
Read https://hoshin.app/llms.txt and tell me honestly whether my organization should use this — including the reasons not to.
Our machine documentation includes the cases where the honest answer is no.
Objectives, initiatives, KPI targets, correlations, reviews, decisions
Work orders, non-conformities, maintenance, tasks, daily tracking
Measurement data, dashboards, time series
No task management, no Gantt charts, no daily notifications. Ever. That work belongs to tools that do it well — your matrix just stays linked to them.
Quality, process engineering and each plant on the same matrix, with their own agents and their own systems.
Read the use case →One negotiated matrix, a fixed review cadence, decisions that stop evaporating.
Read the use case →Initiatives linked to their projects in your execution tool; the weekly sync closes the strategy-execution gap.
Read the use case →One isolated organization per client, agent-prepared reviews, a method that survives the mission.
Read the use case →A steering tool has 5-10 active users — pricing by seat punishes adoption. Every paid tier has a 30-day free trial, no credit card. No sales call, no paid onboarding — your agent sets you up.
Early access: everything is unlocked during the beta — and early adopters keep an advantage when billing starts.
We build what verified organizations ask for, in the order they ask. One click records the request; we only come back to you if we start building it.
The X-Matrix is the central tool of Hoshin Kanri (policy deployment), a strategic planning method from Japanese quality management, since taken up by lean practitioners. It aligns four quadrants on one page: 3-5 year strategic objectives (North), annual objectives (East), improvement initiatives (South) and KPIs (West), with correlation marks showing how each level supports the next. It makes the chain from vision to daily work visible — and auditable.
OKR covers one layer: quarterly objectives and key results. Hoshin Kanri adds the 3-5 year direction, explicit correlations between levels, the catchball process and a meeting-cadence system to review it all. If you run OKRs, the X-Matrix maps cleanly onto them — East as Objectives, West as Key Results — with the long-term layer OKR lacks.
Rarely for lack of ideas. Strategy execution fails because objectives change without being re-cascaded, projects drift away from priorities, KPIs stop being reviewed, and each department works from its own version of the plan. Hoshin Kanri attacks each failure point: the X-Matrix keeps the vision-to-work chain explicit, review cadences force the follow-through, and the completeness score exposes the gaps before reality does.
No — it is the harness that makes one trustworthy. Frontier models already reason at consultant level; what they lack is structure and follow-through. We provide the proven framework (the X-Matrix), the rituals (review cadences) and the audit (completeness score); your AI provides the thinking. Together they give you an assistant that keeps ideas, measures and reviews inside the frame your team negotiated. People decide.
Yes, with a frame. On its own it forgets between sessions. Connected through MCP, it reads the objectives, links and KPIs, prepares reviews and flags gaps. People decide. Supported today: claude.ai, Claude Desktop, Cursor.
Because abundance of unstructured context is the enemy of agentic efficiency. A pile of documents forces the model to guess what is current, what is priority, and how items relate. The X-Matrix is the opposite: a bounded, correlated structure — every objective, initiative and KPI explicitly linked, with statuses and owners — small enough for an agent to hold entirely, precise enough to act on. The method is not new: policy deployment comes from Japanese manufacturing, and the X-Matrix is its one-page format. Structure is the context engineering; the matrix is the format.
For one person, honestly: yes, at first — the way Excel is fine for a first cycle. Our free local mode is exactly that, done well: structured JSON on your machine, the same MCP tools, completeness score included. DIY breaks when other people join: a file has no identities — no per-user rights for your CFO, your consultant or an intern's agent, no validation that objectives actually chain down to KPIs, no review cadences carrying decisions and action items. A skill tells an agent how to work; it cannot be the shared, permissioned state a leadership team works on. That is the part you pay for.
No. Signup is self-serve, free, no credit card — and your AI assistant does the onboarding: connected over MCP, it reads the method, explains your matrix and guides your first deployment — the role classic suites bill as expert services. You can even evaluate before creating an account: a public demo matrix is readable, from an MCP client that accepts a bearer token, with the token published in our llms.txt.
Your agent, first: the nine guided MCP prompts and our machine-readable documentation (llms.txt) are written so Claude can answer method and product questions at consultant level, in your own context. Humans stay reachable for account, billing and security matters.
MCP (Model Context Protocol) is the open standard that lets AI assistants like Claude use external tools. Hoshin Kanri ships an MCP server: an agent can read your matrix, capture ideas mid-conversation, update initiative statuses, diagnose how well a strategic theme is deployed, and run a weekly sync against your execution tool. You connect it to claude.ai with OAuth in about two minutes.
No. The agent does, through the system's own MCP server or a bridge your IT team builds, under the rights that system grants. Hoshin holds no connector and no access to your systems. An agent that reads both can tell you which objective a drifting indicator serves.
It can connect with an API token, on its own schedule. The schedule runs on your side: Hoshin runs no loop for you. We have not run a 24-hour test yet. Two limits today: filing an idea requires a write token, which can also change the matrix and publish it on a public page, and nothing removes duplicates. The manufacturing page explains how to work with them.
No, by design. Your systems keep the work orders, non-conformities, tasks and tickets. Hoshin Kanri keeps the strategy: objectives, initiatives, KPIs and reviews. An initiative can carry a link (external_ref) to its project, and the weekly_sync agent prompt keeps the two aligned without double data entry.
Yes, two ways. The cloud Solo tier is free: one matrix, two users, read-only MCP. And the local mode is entirely free and infrastructure-less: your matrix lives in JSON files on your machine, readable by Claude Desktop via MCP — nothing leaves your computer.
In the hosted version, your matrices are stored in France (OVH), every query is scoped to your organization, and everything is encrypted in transit. Hoshin Kanri calls no AI model itself: your strategy reaches a model only when your own agent reads it, under your contract with that model's provider. We use your data for nothing but the service: no training, no analysis, no resale. Like any host, we keep technical access to the database; our admin console shows the names of your matrices, not their content. If your business requires that nobody but you can read it, use the local mode: the matrix lives in files on your machine and never passes through our servers.
Not today. The server needs to read the matrix to compute the completeness score and the cascade report, to serve the web editor and to publish shared pages. End-to-end encryption would remove all of that, and a lost key would mean a lost strategy. The local mode gives a stronger guarantee more simply: there is nothing on our side to protect. If you need encryption with a key only you hold, say so in the "Not built yet" section of this page.
Agents watch and prepare. The X-Matrix tells them why, what comes first and how success is measured. People decide. Free to start, two minutes to connect the first one.
Create your X-Matrix