Who is this AI worker?
Manage roles, skills, evaluations, and history independently from a model, process, or session.
AI WORKFORCE INFRASTRUCTURE
Move from using AI
Move from using AI
to operating an AI workforce.
Strassen builds the control plane for AI workforces — selecting the right worker, connecting enterprise context, executing safely, and proving the outcome.
IdentityManaged as talent, not as a model
Runtime TruthThe exact runtime, verified before delivery
AuthorityLeast privilege and approval boundaries
Verified OutcomeResult, persistence, and restart proven
THE MISSING LAYER
Agents multiplied.
Workforce management did not.
Enterprises are adopting Claude, Codex, open-source models, internal bots, and specialized agents at extraordinary speed. Yet people still manage the critical questions by hand: who should do the work, where it should run, what it may access, and what evidence proves the job is actually done.
Who should do this work, and where?
Route by quality, cost, privacy, availability, project, and execution environment.
What is the worker allowed to do?
Enforce policy, approval, liveness, generation, and fencing at execution time.
Did the work truly finish?
Prove the business outcome, artifact, persistence, and post-restart read-back — not merely a response or exit code.
FORWARD-DEPLOYED AI ENGINEERING
We don't stop at a proposal.
We go on site and leave a working system.
Strassen engineers work inside the customer's environment to understand the real workflow, data, authority, and success criteria. From there we design AI workers, MCP, Agent Skills, evaluation, approval, and audit as one system, and we own production deployment and outcome verification — not a PoC. Common problems found in the field flow back into hiyu and reusable implementation patterns.
Start with the work.
Not with model selection: with the actual workflow, its bottlenecks, its owner, and its success metrics.
Build in context.
Ship something that runs inside the customer's data, permissions, existing systems, and security requirements.
Prove the outcome.
Not by how a demo looks: by completion rate, quality, time, cost, exception handling, and operational continuity.
HIYU
The control plane
for AI workforces.
hiyu registers, evaluates, selects, and assigns AI workers across local models, cloud models, CLIs, software systems, and physical resources. It does more than dispatch instructions. It verifies that the intended worker acts with the intended authority, in the intended environment, and that the result survives reality.
TalentBankpersona · role · skill · evidence · fit
Routingmodel · account · cost · privacy · project · cwd
Policy & Approvalleast privilege · human decision · audit
Safe Deliveryexact process · generation · fencing · quarantine
Durable Scheduleroccurrence · run · attempt · retry · artifact
Repair & Evaluationevidence · proposal · validation · atomic apply
Completion & Auditlive outcome · persistence · restart · read-back
Fleetlocal · cloud · CLI · software · physical workers
Cloud ModelsLocal LLMsCLI AgentsSoftware SystemsPhysical Workers
FROM INTENT TO EVIDENCE
Close every request
with a verifiable outcome.
01
Intent & Contract
goal · scope · acceptance
02
Talent & Route
who · where · which runtime
03
Policy & Delivery
authority · approval · liveness · fence
04
Execution
real system actions · result · artifact
05
Verify & Persist
independent verification · business outcome · persistence
06
Restart & Read-back
controlled restart · recovery · durability
VERIFIED COMPLETE
hiyu does not treat “the AI replied” as success. Work is complete only when the intended effect occurred in the real system, the result was persisted, and the state can be read back after restart.
NON-NEGOTIABLE
Four truths for operating
AI in production.
MODEL ≠ TALENT
Models are replaceable resources. An AI worker's identity, role, evaluation, and history must remain independent from the current model.
GREEN DOT ≠ ALIVE
A registry entry or terminal pane can remain after the provider process has exited. Verify the exact runtime immediately before delivery.
SCHEDULED ≠ EXECUTED
Saving a cron expression is not business execution. Inputs, authority, placement, artifacts, retries, results, and restart recovery belong to the scheduler contract.
MERGED ≠ COMPLETE
Tests, pull requests, merges, and deployments are intermediate states. Completion requires a live outcome and authoritative read-back.
ONE SYSTEM, MULTIPLE SURFACES
Not one giant application.
A system that lets AI work.
CONTROL PLANE · FLAGSHIPBuilding in Public
hiyu
AI Workforce Control Plane
Select AI talent by role and evidence, assign the effective runtime, and manage work through verified completion. The name, 日結, means tying people and AI together for daily results.
Explore hiyuCONTEXTPublic Pilot
UDQ
Private Enterprise Context
Turn contracts, meeting notes, policies, and proposals into grounded answers and context for AI workers — without moving the source documents outside the chosen boundary.
Explore UDQCOLLABORATIONAvailable · Team plan
engawa
Collaboration & Memory
Turn conversations, decisions, and outcomes shared by people and AI into organizational memory and future work. Named after the engawa (縁側), the veranda where people naturally gather.
Explore engawaINTAKEResearch / In Development
ticket-intake
Words to Work Contracts
Translate an ambiguous request into a work contract with goals, scope, authority, acceptance criteria, and required artifacts.
Product status is kept in step with the availability stated on each product site.
STRASSEN FORWARD LOOP
Start with production,
not another demo.
Enterprise AI usually stalls not because the model is incapable, but because data, authority, legacy systems, exceptions, and ownership never connect in the real environment. Strassen closes one workflow in production, then turns the deployment into reusable connectors, skills, policies, evaluations, and runbooks.
SkillsConnectorsEvalsPoliciesRunbooks
Discover
Whose work, which workflow, how far. Baseline, success metrics, ownership, and approval points.
Ground
Source of truth, freshness, sensitivity, identity and authorization, and the boundary between retrieval and write actions — as a contract.
Build
AI worker roles, tool contracts, MCP, Skills, deterministic validators, human approval, failure recovery.
Verify
Completion rate, correctness, blocked permission violations, escalation, regression, restart durability, read-back.
Deploy
Exact candidate version, production route, monitoring and alerts, rollback, runbook, post-restart state check.
Compound
Skills, connectors, eval suites, security patterns, playbooks, and hiyu requirements — so the next workflow ships faster.
PROOF, NOT PROMISES
The larger the claim,
the more precise the evidence.
Evidence Contract
Strassen separates prototypes, offline tests, production behavior, and customer outcomes.
Every published metric should carry its scope, environment, baseline, measurement window, and verification date. What has not been measured remains explicitly unmeasured. This page does not yet publish verified production metrics.
AProduction outcomeactual user route
BReproducible evidenceinstructions + artifacts
CControlled validationfixed environment
DRoadmap / conceptexplicitly labeled
DESIGN PHILOSOPHY
Systems where AI is the protagonist.
Strassen's mission is to redefine systems around AI — to build a society where AI does the work. Most companies aim to make internal data usable by AI. Strassen goes one step further: we redefine the system itself so that AI is the principal actor. If AI can carry a job through to the end, people should not even need to operate a chat or a UI — that is what we design toward. The human role is to hand over intent, set the boundaries, and approve the result.
The foundation AI uses directly.
Primitives for AI teams to create, embed, and retrieve knowledge. Designed for direct use by AI, and to run standalone — without the cloud service or a human-facing UI.
A layer for people.
A wrapper that makes the core usable by humans. It exists for convenience; it is not a precondition for AI to function. If AI does everything, this layer is not essential.
hiyu's structure follows from this philosophy. This describes design intent; availability follows each product's status.
OUR THESIS
The enterprise will not buy one AI agent.
It will operate a workforce.
Models will become more capable, cheaper, and interchangeable. Durable advantage will move to enterprise context, execution routes, authority, evaluation, and proven outcomes. Strassen is building that operating layer.
IDENTITYCONTEXTROUTINGAUTHORITYOUTCOME
THE SYSTEMS BEHIND AI WORK
We publish the failure conditions,
not only the success stories.
Technical articles in preparation — each with a problem statement, failure model, invariant, architecture, and either evidence or an explicit NOT MEASURED.
PR merged is not complete
Completion Contracts for AI work. Read (Japanese) →
AI worker identity is not a session
Separate talent from runtime.
When the AI exits but the shell stays alive
Safe delivery at the process boundary.
Cost per verified completion
A better unit than cost per token.
TOKYO / HAMAMATSUCHO
Built in Tokyo.
Deployed where the work happens.
Strassen develops products and deployment methods that connect AI to real work. We welcome technical conversations with enterprises, startups, investors, and builders.
5F, Saison Bldg. Hamamatsucho1-23-9 Hamamatsucho, Minato-ku
Tokyo 105-0013, JapanApproximately one minute from Daimon Station (Toei Asakusa / Oedo lines) and two minutes from JR Hamamatsucho Station.Open in Maps ↗ · Visits are by appointment.
大門 / DaimonS浜松町 / Hamamatsucho
- Company
- Strassen Inc. (株式会社Strassen)
- Business
- AI Workforce Infrastructure (hiyu / UDQ / engawa) and Deployment Engineering
- Representative Director
- Osamu Shibata
- Address
- 5F, Saison Bldg. Hamamatsucho, 1-23-9 Hamamatsucho, Minato-ku, Tokyo 105-0013, Japan
- Contact
- Contact form
BRING ONE WORKFLOW
Bring us one workflow
that refuses to leave pilot mode.
We will connect the model, company context, authority, execution, verification, and ongoing operation inside the real environment. Three routes, depending on what you need.