Matrix Land SI feature guide

Build applications, retain methods, find information and delegate tasks. Nine modules explain how Matrix Land SI supports everyday work.

Start with what you want to do

To build a team tool, start with App. To retain a working method, look at Skill. To organize shared information, use Knowledge. To delegate ongoing work to an assistant, look at Mate. Connection, development and model services support these activities.

Business work: App · Skill · Knowledge · Mate
Connections and identity: MCP · CLI · Auth SDK
Shared support: Token Service · Community

These are reading groups. A real task may cross several: a Mate can read a Skill, query an application and Knowledge as an employee, then call a model to organize the results.

For each module, first read the problem it solves and its capabilities, then how it works and connects to the others. Examples explain working methods; actual applications still need configuration for their business requirements.

01LEGO APP

A creation platform for business people

Turn something you do every day into an application your team can actually use.

What problem it solves

Many business needs sound small: following up with customers, organizing event registrations or checking expenses. A spreadsheet can get things started, but more people require shared data, clear permissions and a stable entry point. AI can already help business people write programs. LEGO addresses what follows: where the program runs, where data lives, who uses it and how it changes later.

Capabilities

Applications and runtime space: Register an application’s name, owner and entry point, and host webpages or business programs. An online manual can be an application, as can a customer-follow-up tool.
Data, files and background work: Configure databases, file storage, lightweight caches, task queues and scheduled triggers as needed, or connect existing databases. Business records, attachments and time-consuming work each get an appropriate place.
Versions and releases: Retain source snapshots, runtime artifacts and release notes. A prepared version becomes available to production users only after publication. Earlier code versions can be selected again when needed.
Business capabilities shared by people and AI: People operate the webpage; an application can also declare MCP tools so agents query or modify business objects under its rules. Login, application access and record-level permissions are determined separately.

How it works

Business rules
Application code and resources
Prepare a version
Release to users

Define the business first

For a customer-follow-up tool, define what a follow-up record is, who owns it and which status transitions are allowed. Business people supply these judgments; AI and developers turn them into interfaces, data structures and operating rules. LEGO provides runtime resources.

Turn changes into an explicit version

Code is completed and built in a development environment, then source and runtime artifacts are uploaded through the CLI. LEGO records the version’s configuration and resources. Publishing switches the production entry point to the selected version, rather than treating a saved file still under development as a release.

Let both entry points operate on the same facts

An employee updates status on the webpage while Mate queries progress through application tools; both ultimately use the same business record handled by the application. Those tools must be explicitly provided. Building a webpage does not automatically create AI-callable business interfaces.

Its place in Matrix Land SI

LEGO handles application delivery; Auth handles identity and access; CLI connects local development to the cloud; MCP lets agents use defined business operations.

02LEGO Skill

Keep a working method for next time

The useful thing is not only an answer, but the method that produced it.

What problem it solves

The first time a colleague asks AI to summarize a meeting, they often explain what matters, how to separate decisions from suggestions and how detailed action items should be. A new conversation requires the explanation again. Skill retains stable, repeatable methods so the employee, colleagues or Mate can reuse them for similar tasks.

Capabilities

Method instructions: Use SKILL.md to explain when to use the method, what materials it needs, how to work and what the output must satisfy. It can hold a department’s accumulated experience and judgment criteria.
Supporting materials: Keep templates, examples and necessary supporting files in the same Skill. Instructions explain the method; materials make it executable and provide references.
Version management: Create a new version when improving a method and preserve its provenance. Mate can reference a specific published version, giving formal work a verifiable methodological basis.
Sharing and reuse: Make methods discoverable through appropriate sharing or Skills Hub. One method may support different tasks and be used by different Mates.

How it works

Practical experience
Method and examples
Skill version
Agent adopts as needed

Extract the method from real work

Begin with something that has already succeeded. Good meeting notes, for example, separate conclusions, disagreements, owners and deadlines. Writing those judgments into a Skill with a good example produces more consistent results than merely asking for a “professional summary.”

Read the method when needed

A Mate’s published configuration links to Skill versions, whose relevant instructions are read at runtime according to the task. Methods can then be maintained independently, without putting every professional detail into every conversation. Task-specific materials are obtained during execution.

Maintain methods and information separately

Meeting-note methods belong in a Skill; this week’s raw meeting materials belong in Knowledge; official action-item status belongs in an application. Their update frequencies differ, making it easier to locate what actually needs changing when they are separate.

Its place in Matrix Land SI

Skill stores methods; Knowledge stores reference information; Mate applies methods to tasks; Community helps people discover reusable methods.

03Knowledge

Give information provenance and an owner

Before answering, find the document that should support the answer.

What problem it solves

Company information is scattered across chat attachments, personal folders and copies. Fluent AI answers may still cite obsolete rules or miss conditions. Knowledge manages files, directories, versions and readership together, so employees and agents find the same materials and know where conclusions come from.

Capabilities

Organization and reading: Upload files, create directories, rename and move content. Read or preview formats including text, Markdown, HTML, images and PDFs, and import ZIP archives.
Versions and collaborative editing: Retain historical content and support updates, appends and patches. Restoring old content creates a new current version. Editing checks the version to avoid overwriting someone else’s recent changes.
Search and knowledge services: Ordinary folders support listings, search and precise lookup. Enabling Knowledge Service adds scope descriptions, a Handbook and indexed retrieval.
Access and feedback: Manage readers and collaborators by root folder, with explicit sharing handled separately. Users can report problems for maintainers to evaluate and correct in the source.

How it works

Maintain original files
Directory and scope description
Search and locate
Read back and cite

Source files, directories and indexes each have a role

Source files retain content, directories organize it and indexes help locate it. A Handbook tells agents what is available, which questions it suits and how to search. Search results are clues to evidence; important conclusions require reading the underlying file.

Organize information by actual readership

Contents under one root folder inherit its permissions. Company-wide policies and team-only materials belong in separate roots. A knowledge service adds ways to query; readership still needs separate definition.

Keep maintenance aligned with use

After updates, source files and search indexes may temporarily differ. Important questions require checking dates, versions and applicability. A PDF or image being uploadable or previewable does not establish that searchable text exists or that AI has understood all its contents.

Its place in Matrix Land SI

Knowledge retains traceable information; Auth supplies employee identity; Mate and other agents search within the user’s permissions; business applications retain live business records.

04Mate

Give AI responsibilities for ongoing work

Beyond answering this message, know what the job requires next.

What problem it solves

General chat suits temporary discussion. Ongoing work also needs stable responsibilities, tools, methods and execution records. Mate combines these in one assistant: personal secretaries help with daily affairs, while specialist Mates serve particular kinds of work. Employees need not re-explain the role and requirements every time.

Capabilities

Role and working instructions: SOUL describes who it is and its responsibilities; Handbook describes how it works; Skill supplies reusable methods. Model, resource and entry-point settings together form a publishable definition.
Conversations and task execution: Understand requests in a session, call available tools, retrieve information, process tasks and deliver results. Execution may produce files, operations awaiting confirmation and follow-up reminders.
Specialization: Personal secretaries and specialist Mates can handle different work. Choose a Mate by its responsibilities and available capabilities, not merely its name or avatar.
Preview and production versions: Preview an edited definition before publishing an explicit version. Production uses published configuration, making changes to roles, methods and capabilities traceable.

How it works

Responsibilities and methods
Publish Mate
Receive employee tasks
Execute, confirm and deliver

Define the work, then hand it to the runtime

LEGO stores Mate definitions and published versions; Runtime handles actual sessions and tasks. The former defines how the assistant should work; the latter calls models and tools and records progress within a specific task.

Retrieve methods and facts around the task

Mate selects Skills from its instructions, finds evidence in Knowledge and reads business status from applications. Preparing a meeting can combine a note-taking method, previous minutes and current project progress without asking the user to paste everything into chat.

Act as the current employee

Mate’s resource access is constrained by the actual caller’s authorization, rather than inheriting all of its creator’s permissions. Actions requiring confirmation enter the relevant approval process. A conversational suggestion does not automatically rewrite formal instructions; lasting rules require explicit maintenance and publication.

Its place in Matrix Land SI

Mate brings models, methods, knowledge and business tools into one job. LEGO manages definitions, Runtime manages execution, and Auth and resource services jointly enforce access rules.

05Workspace MCP

Give agents an interface they can understand

Make clear what AI can do, which parameters it needs and what it receives afterward.

What problem it solves

People can inspect webpages and find buttons; agents are better served by clearly defined tools. If every system connects differently, each new assistant requires another integration. Workspace MCP provides a unified entry point for platform resources, allowing compatible agents to discover capabilities, read information and operate within authorization.

Capabilities

Capability discovery: Describe operations with tool names, purposes and parameters. Agents can inspect tools before selecting an action appropriate to the task.
Platform resource operations: Connect online resources including Knowledge, LEGO applications and release records, and Skills. Tool definitions and current authorization jointly determine the callable scope.
Application business tools: Applications can register query and action interfaces with required roles and parameters, letting agents work directly with business objects.
Identity and result feedback: Calls carry trusted identity. The platform and target service check permissions and return results or specific errors for the agent’s next decision.

How it works

Agent selects a tool
Verify caller identity
Resource checks and execution
Return results

Express a business action as a contract

“Update project status,” for example, should define the project identifier, permitted new states, who may act and what is returned. The tool definition explains the request; the application code that changes the record performs validation and execution.

Check authorization on every call

Agents authenticate and obtain authorization before calling tools. The unified entry point checks calling eligibility; the target service checks the specific resource and business rules. Discovering a tool means knowing it exists. Permission to act on an object still depends on the request.

Connect online resources with local files

MCP can read online applications and knowledge resources but cannot directly see project files on a developer’s computer. After the CLI uploads local build outputs, an agent can inspect versions or publish through online tools. The two tools handle work in different locations.

Its place in Matrix Land SI

MCP unifies calls between agents and online capabilities; Auth confirms identity; each service retains its data and enforces rules; CLI delivers local files.

06Workspace MCP CLI

Connect projects on your computer to the cloud platform

Code is written locally and applications run online. They need an explicit delivery path between them.

What problem it solves

AI coding tools edit files, run checks and build projects on a computer. Platform MCP manages online resources. CLI connects those environments so local agents can sign into Matrix Land SI, use platform capabilities and deliver actual files to LEGO. Business people can give AI the goal while developers retain familiar local workflows.

Capabilities

Login and connection: Establish a Matrix Land SI identity connection for local tools and provide access to platform tool calls. Development can use a unified account without copying credentials throughout projects.
Source snapshots: Back up selected project source into application records to retain code provenance. Snapshots may be saved independently for later inspection and handover.
Build artifact delivery: Upload runnable build outputs with a version number, name and upgrade notes, preparing an application version for release.
Database schema migration: Read local migration directories, plan and verify changes, then apply permitted schema updates. Data structures required by code receive a separate management process.

How it works

Local editing and building
CLI upload
LEGO records the version
MCP publication

Create and check locally

An editor or coding agent modifies source, and the project’s build tools turn it into runnable files. CLI receives and delivers those files to the platform. Different stacks can therefore use the development tools appropriate to them.

Retain both provenance and the runnable version

Source explains how the application was made; build artifacts contain what actually runs. Link both to one version record so later maintainers can identify which changes produced the online application and take over the work.

Separate preparation, publication and data changes

A successful upload produces a version awaiting release; an online publication operation performs the production switch. Database migration has its own plan and result. Even when code returns to an earlier version, executed data changes need separate treatment. This makes the object of each change clear.

Its place in Matrix Land SI

CLI handles local files, login connections and delivery; LEGO retains application versions and resource relationships; MCP handles subsequent online operations. Together they connect AI coding to production use.

07Workspace Auth SDK

Help applications recognize employees and distinguish permissions

What problem it solves

Rebuilding accounts, employee lists and login flows for each application can give one person inconsistent identities across systems. Workspace Auth SDK connects applications to unified identity and associated authorization capabilities. Applications focus on business rules while employees use their company identities.

Capabilities

Unified login integration: Coordinate browser and server login capabilities, establishing application sessions from trusted authentication results and reducing duplicated identity work.
Employee and department directory: Use unified employee and department identifiers within authorized limits. Owners, participants and departments in applications can refer to consistent identities.
Resource and tool authorization: Support credential validation for target resources and role declarations and checks for application MCP tools, carrying identity into actual operations.
A development package used with the application: The official SDK is delivered in explicit versions for developers. It becomes part of application code; upgrading requires updating the dependency and validating the application’s behavior.

How it works

Unified employee identity
Trusted application session
Application and role authorization
Specific record checks

Establish sessions from trusted login results

After company login, the application backend verifies the callback and establishes its own session. Subsequent requests obtain identity from that session. A name or employee number entered on a page is only input, not verified login identity.

Distinguish three questions

Unified identity answers “Who are you?” Application access settings answer “May you enter?” The business backend must still answer “May you change this record?” Entering a project system, for example, does not permit changing every project’s owner.

Enforce rules where actions execute

Interfaces may show features by role, but backends must still check each operation. MCP calls follow the same principle: tools declare required roles, while execution code checks objects and data scope. A role name alone does not establish complete business permissions.

Its place in Matrix Land SI

Auth provides trusted identity and access capabilities; SDK helps applications adopt them; LEGO registers applications; applications remain responsible for access rules on their business data.

08Token Service

Use models consistently and see consumption clearly

Make models available to applications and show teams where they are being used.

What problem it solves

Applications and agents all need models. Separate vendor accounts and scattered keys quickly complicate integration and cost management. Token Service provides unified model access and usage management. Developers use permitted models through the platform; administrators understand consumption and budgets by purpose.

Capabilities

Manage keys by purpose: Request and manage API keys for agents, applications and other uses, giving credentials explicit ownership for later management and revocation.
Model catalog: Inspect available models, capabilities and pricing to choose one suited to the task. Actual availability and prices follow the catalog at the time.
Budgets and usage: Inspect key-level consumption and budget constraints, helping distinguish development experiments, personal use and application service costs.
Unified calling entry point: Access vendor services through a model gateway, supported by integration examples and runtime status information, reducing the need for each application to maintain separate upstream connections.

How it works

Application or agent
Purpose-specific key
Model gateway
Results and usage records

Centralize model integration

Applications and agents request models using platform-assigned credentials. The gateway connects configured services and records usage. Upstream vendor connections are managed on the platform, while applications focus on their inputs, outputs and business processing.

Choose models by task requirements

Simple text classification and an analysis requiring multiple rounds of refinement have different capability, latency and cost needs. Test real samples to determine whether results are sufficient, then consider observed usage. A model name cannot replace task validation.

Manage computation separately from business authorization

A model key determines whether a model can be called and where consumption is charged. Customer information, company files and application operations still need authorization from their business services. Applications should use protected credentials in trusted backends and deliver results through application features.

Its place in Matrix Land SI

Token Service manages model access and consumption; LEGO App, Mate and external agents use the models; Auth and resource services manage business permissions. Independent tools such as search and image services connect through their own interfaces and cannot be inferred from the model catalog alone.

09Community

Turn one person’s practice into experience others can use

After making something work, help others find it, understand it and continue improving it.

What problem it solves

AI lowers the barrier to tool creation but also brings repeated experiments. Several departments may explore the same spreadsheet method, while a useful small application is known only to its author’s neighbors. Community connects discussion, experience sharing and reusable results so others can build on existing practice.

Capabilities

Discussion and exchange: Discuss needs, usage experiences and concrete problems, helping authors understand real difficulties and users understand the judgments behind methods.
Skills Hub and application discovery: Find reusable Skills and Apps by purpose, description, provenance and version, then follow through to their content or application entry points.
Submission and review: Submit prepared work for sharing. Review produces a listing that explains what is shared, which version it comes from and whom it suits.
Versions and ongoing maintenance: Shared listings link to original content or application records. Methods, descriptions and versions can continue changing so discovery keeps pace with maintenance.

How it works

Practice at work
Description and submission
Review and discovery
Feedback and new versions

Explain applicability first

When sharing a Skill, describe the problem it has solved, required inputs and expected output. When sharing an application, identify its intended users and maintainer. Others can then judge its fit rather than see only a name.

Make listings point to explicit sources

Marketplace entries store introductions, review and display information, while Skill contents and App versions retain their own sources. People discover work through Community and follow those sources to content, runtime entry points and maintenance relationships.

Bring feedback into maintenance

Maintainers decide whether new requirements and counterexamples from discussion fit the original purpose. Accepted changes update content or an application version, followed by the shared description. A discussion does not automatically rewrite everyone’s tools; formal improvements must reach an actual version.

Its place in Matrix Land SI

Community supports discovery and exchange; LEGO retains Skill and application provenance; Auth handles access rules. Listings, code releases and business permissions each have records. Finding an application listing does not change the access limits governing its use.