{
  "schema": "lazurio.documentation.content-index.v1",
  "canonicalBaseUrl": "https://documentation.lazurio.ai",
  "sourceRepository": "https://github.com/Lazurio/documentation",
  "sourceCommit": "5c210aa8adfaa362fa0c4b2046c12c8742861e20",
  "sourceTreeDirty": false,
  "generatedAt": "2026-08-26T22:24:16+02:00",
  "sources": [
    {
      "id": "lazurio-readme",
      "title": "Lazurio README",
      "url": "https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/README.md",
      "publisher": "Lazurio"
    },
    {
      "id": "lazurio-architecture",
      "title": "Lazurio architecture",
      "url": "https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/ARCHITECTURE.md",
      "publisher": "Lazurio"
    },
    {
      "id": "lazurio-collaboration-model",
      "title": "Lazurio collaboration and access model",
      "url": "https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/AGENTS.md",
      "publisher": "Lazurio"
    },
    {
      "id": "lazurio-external-apps",
      "title": "External application integration standard",
      "url": "https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/manual/external-app-integrations.md",
      "publisher": "Lazurio"
    },
    {
      "id": "lazurio-secret-custody",
      "title": "Local secret custody standard",
      "url": "https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/manual/security/local-secret-custody.md",
      "publisher": "Lazurio"
    },
    {
      "id": "microsoft-copilot-architecture",
      "title": "Microsoft 365 Copilot architecture and how it works",
      "url": "https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-architecture",
      "publisher": "Microsoft Learn"
    },
    {
      "id": "microsoft-copilot-data-protection",
      "title": "Data protection and auditing for Microsoft 365 Copilot",
      "url": "https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-architecture-data-protection-auditing",
      "publisher": "Microsoft Learn"
    },
    {
      "id": "microsoft-copilot-privacy",
      "title": "Data, privacy, and security for Microsoft 365 Copilot",
      "url": "https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-privacy",
      "publisher": "Microsoft Learn"
    },
    {
      "id": "microsoft-copilot-requirements",
      "title": "Microsoft 365 Copilot requirements",
      "url": "https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-minimum-requirements",
      "publisher": "Microsoft Learn"
    },
    {
      "id": "microsoft-copilot-extensibility",
      "title": "Microsoft 365 Copilot extensibility",
      "url": "https://learn.microsoft.com/en-us/microsoft-365/copilot/extensibility/",
      "publisher": "Microsoft Learn"
    }
  ],
  "documents": [
    {
      "stableId": "lazurio-doc-agents",
      "locale": "en",
      "route": "/en/agents/",
      "canonicalUrl": "https://documentation.lazurio.ai/en/agents/",
      "sourcePath": "data/v2/docs/en/agents.md",
      "title": "Documentation for agents",
      "summary": "Use llms.txt and the content index to discover reviewed Lazurio pages, evidence links, audiences, review dates, and stable identities.",
      "audience": [
        "builder",
        "agent"
      ],
      "updatedAt": "2026-08-26",
      "reviewedAt": "2026-08-26",
      "reviewOwner": "Matej Suchanek",
      "secondReviewOwner": null,
      "trustCritical": false,
      "sourceRefs": [
        "lazurio-readme",
        "lazurio-architecture"
      ],
      "markdown": "Agents should read the same reviewed source as people. This site does not keep\na hidden “AI version” of Lazurio documentation.\n\n## Discovery surfaces\n\n- [`/llms.txt`](/llms.txt) is a compact map of canonical pages and summaries.\n- [`/content-index.json`](/content-index.json) is the versioned structured\n  index. It includes stable document IDs, routes, audiences, review metadata,\n  canonical URLs, public source paths, evidence references and Markdown content.\n- [`/sitemap-index.xml`](/sitemap-index.xml) is the web crawler map generated\n  by the documentation build.\n- Human-readable pages remain the canonical interpretation surface.\n\nStable-ID retrieval defaults to the English document. Agents can request a\ndifferent published locale explicitly; canonical routes always select the exact\nlocalized page.\n\nAll artifacts are built from `data/v2/docs` in the public\n[documentation repository](https://github.com/Lazurio/documentation). The\nindex identifies the exact source commit and whether it came from a dirty local\npreview. Production deployment rejects a dirty or non-approved commit.\n\n## Retrieval guidance\n\n1. Start with `llms.txt` to select a page.\n2. Prefer a stable ID from `content-index.json` when persisting a reference;\n   routes may be localized later.\n3. Read `sourceRefs` and the matching public source records before repeating a\n   trust-critical claim.\n4. Distinguish documented provider facts from Lazurio's assessment.\n5. Report the page review date and source commit when freshness matters.\n6. If a statement is absent or deployment-specific, say so instead of filling\n   the gap from inference.\n\n## Future MCP server\n\nA later read-only MCP server may expose discovery and retrieval over the same\ncontent index. It will not become a writer, private overlay or second content\nstore. The current JSON contract is the compatibility seam for that work; the\nwebsite does not depend on the future server."
    },
    {
      "stableId": "lazurio-doc-data-access-security",
      "locale": "en",
      "route": "/en/data-access-security/",
      "canonicalUrl": "https://documentation.lazurio.ai/en/data-access-security/",
      "sourcePath": "data/v2/docs/en/data-access-security.md",
      "title": "Data access and security",
      "summary": "Review identity, Organization isolation, local files, integrations, secrets, model providers, audit evidence, and residual risks.",
      "audience": [
        "it-admin",
        "decision-maker",
        "agent"
      ],
      "updatedAt": "2026-08-26",
      "reviewedAt": "2026-08-26",
      "reviewOwner": "Matej Suchanek",
      "secondReviewOwner": "Pablo AI",
      "trustCritical": true,
      "sourceRefs": [
        "lazurio-architecture",
        "lazurio-collaboration-model",
        "lazurio-external-apps",
        "lazurio-secret-custody"
      ],
      "markdown": "Lazurio's security model starts with a deliberately limited claim: an agent\ncannot be made safe by prompt wording alone. Security comes from the identity,\nmachine, repository permissions, scoped tools, secret custody and publication\ncontrols around the session. The public [collaboration model](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/AGENTS.md)\ndefines the intended boundary; a concrete deployment still needs verification.\n\n## Trust boundaries\n\n### Principal and task agent\n\nThe task agent operates for the signed-in principal. It must not acquire rights\nmerely because a prompt asks it to. Effective access is supplied by the\nprincipal's device sessions, repository grants and provider credentials. This\nmeans offboarding and access review must include those underlying systems.\n\n### Organization\n\nAn Organization is the company-level data and access boundary. Separate\norganizations remain separate repositories and GitHub organizations. Public\npatterns can move between them; secrets, client data, business strategy and\nprivate overlays cannot.\n\n### Personalspace\n\nPersonalspace is private to one principal and is not a shared company data\nstore. Its privacy boundary has priority over convenience. Company knowledge\nneeded by colleagues belongs in an authorized Organization store, not in\nsomeone else's private context.\n\n### External applications\n\nThe [integration standard](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/manual/external-app-integrations.md)\nprefers a locally curated official MCP server, then an official CLI, then a\nreviewed and pinned open-source implementation. Browser interaction is a\nfallback. Each machine uses a separately revocable login, and the deployment's\ncatalogue should contain names and required scopes—not secret values.\n\n### Secrets\n\nThe [secret custody standard](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/manual/security/local-secret-custody.md)\nkeeps real credentials out of Git. Local ignored paths are scoped by owner or\nOrganization, and tracked source contains only schemas, variable names and\ninstructions. Your endpoint and backup controls must protect the actual local\ncustody location.\n\n## Threats to test\n\n| Threat | Intended control | Acceptance test |\n| --- | --- | --- |\n| Prompt requests data from another company | Organization isolation and scoped context | Attempt access with no grant and confirm denial and useful audit evidence. |\n| Agent attempts a protected publication | GitHub/provider permissions plus explicit principal approval | Attempt merge/deploy without required permission or review. |\n| Credential is copied into source | Ignored custody paths, public-safety scanning and review | Seed a safe canary matching a token pattern and confirm the pipeline fails. |\n| Integration has excessive access | Provider-side scopes and separate revocation | Read live OAuth/app grants and revoke one without affecting unrelated access. |\n| Local machine is lost | Device controls, encryption, credential revocation and recovery procedure | Run the organization's offboarding or lost-device exercise. |\n| Generated explanation drifts from behavior | Exact public source links, review dates and CI validation | Change or expire an evidence reference and confirm the documentation build fails. |\n\n## Model-provider boundary\n\nLazurio can be used through different execution clients and model providers.\nThis documentation does not make one universal data-retention or training\nclaim for all of them. The operator must disclose the selected provider,\naccount tier, region where relevant, enabled telemetry, retention terms and any\nzero-data-retention arrangement. Treat those provider terms as part of the\ndeployment—not as an inherited property of Lazurio.\n\n## Audit expectations\n\nGit history, pull requests, exact-head approvals and deployment records provide\nstrong evidence for source changes. They do not automatically cover every\nmodel request, local file read or third-party API call. Build an audit map that\nstates which system records identity, action, target, result and retention for\neach enabled surface.\n\n## Residual risk\n\nAn authorized identity can still expose information, approve a harmful change,\nor grant a tool too much access. A process boundary can be misapplied. A model\ncan generate incorrect content. A local endpoint can be compromised. Lazurio's\nvalue is to make these decisions narrower and more inspectable; it does not\nremove the need for least privilege, endpoint security, provider review,\ntesting and human accountability."
    },
    {
      "stableId": "lazurio-doc-deployment-operations",
      "locale": "en",
      "route": "/en/deployment-operations/",
      "canonicalUrl": "https://documentation.lazurio.ai/en/deployment-operations/",
      "sourcePath": "data/v2/docs/en/deployment-operations.md",
      "title": "Deployment and operations",
      "summary": "Plan a Lazurio rollout across identity, devices, repositories, modules, integrations, logs, backup, updates, and offboarding.",
      "audience": [
        "it-admin",
        "builder",
        "agent"
      ],
      "updatedAt": "2026-08-26",
      "reviewedAt": "2026-08-26",
      "reviewOwner": "Matej Suchanek",
      "secondReviewOwner": "Pablo AI",
      "trustCritical": true,
      "sourceRefs": [
        "lazurio-readme",
        "lazurio-architecture",
        "lazurio-collaboration-model",
        "lazurio-external-apps",
        "lazurio-secret-custody"
      ],
      "markdown": "Lazurio is a framework with local workspaces and independently owned modules,\nnot one universal hosted topology. The public [architecture](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/ARCHITECTURE.md)\ndefines the common model. A rollout record must define the actual devices,\nrepositories, providers and deployed services used by your Organization.\n\n## Components to account for\n\n1. **Principal machine:** the endpoint where a person or AI colleague works.\n2. **Lazurio root:** the local guide and launch surface above authorized\n   Organization checkouts.\n3. **Organization repository:** the company boundary, configuration and\n   shared sources of truth.\n4. **Workspace modules:** applications with their own runtime, dependencies,\n   checks and deployment target.\n5. **GitHub:** repository identity, team grants, review history and branch\n   enforcement in the documented default model.\n6. **Execution client and model provider:** selected by the deployment and\n   reviewed under its own commercial and data terms.\n7. **External applications:** individually enabled MCP servers, CLIs or\n   browser workflows with provider-side scopes.\n\nNot every module must be deployed to a public server. Some are local tools,\nsome are internal services, and some publish a public surface such as this\ndocumentation site. Module ownership keeps those choices explicit.\n\n## Rollout sequence\n\n### 1. Define the Organization\n\nName the GitHub organization, repositories, teams, administrators and\noffboarding owner. Decide what company data belongs there and what is excluded.\n\n### 2. Approve an endpoint baseline\n\nDocument supported operating systems, device ownership, encryption, screen\nlock, patching, endpoint monitoring, local backup, remote wipe and incident\nhandling. A principal-owned machine still needs company-grade controls when it\nprocesses company data.\n\n### 3. Select the execution provider\n\nRecord the client, model provider, account type, authentication, data handling,\nretention, telemetry and contractual owner. Repeat this review when the client\nor account tier changes.\n\n### 4. Enable the minimum repositories and integrations\n\nStart with one bounded use case. Give the principal only the repositories and\nprovider scopes it needs. Follow the public [external application\nstandard](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/manual/external-app-integrations.md)\nand test revocation.\n\n### 5. Enforce publication rules\n\nSet branch protection, required checks and reviewers appropriate to the\nrepository. Define equivalent approval points for messages, infrastructure,\nbilling, secrets and destructive provider actions.\n\n### 6. Run an acceptance exercise\n\nProve the normal task, a denied access attempt, credential revocation, failed\nCI, rollback, offboarding and incident escalation. Keep evidence with the\nrollout decision.\n\n## Updating and rollback\n\nLazurio source, Organization configuration and each module are versioned\nindependently. Updates should fast-forward clean primary checkouts, run the\ndeclared doctor/check gates and enter production through a reviewed exact\ncommit. Rollback means returning the affected repository or deployment to a\nprevious verified revision; it must not silently restore revoked credentials\nor superseded access.\n\n## Operating questions that remain deployment-specific\n\n- Who maintains endpoints, GitHub teams and external integrations?\n- Which model provider and account terms apply?\n- Where do module services run, and which networks can reach them?\n- Which logs exist and how long are they retained?\n- How are local data and credentials backed up, deleted and recovered?\n- What response time is promised, if any?\n\nThese answers should be part of the customer-specific acceptance package.\nThis public documentation intentionally does not invent them."
    },
    {
      "stableId": "lazurio-doc-faq",
      "locale": "en",
      "route": "/en/faq/",
      "canonicalUrl": "https://documentation.lazurio.ai/en/faq/",
      "sourcePath": "data/v2/docs/en/faq.md",
      "title": "Frequently asked questions",
      "summary": "Answers about what Lazurio is, whether it replaces Copilot, data access, local deployment, approval, audits, and the future MCP server.",
      "audience": [
        "it-admin",
        "decision-maker",
        "builder",
        "agent"
      ],
      "updatedAt": "2026-08-26",
      "reviewedAt": "2026-08-26",
      "reviewOwner": "Matej Suchanek",
      "secondReviewOwner": "Pablo AI",
      "trustCritical": true,
      "sourceRefs": [
        "lazurio-readme",
        "lazurio-architecture",
        "lazurio-collaboration-model",
        "lazurio-external-apps",
        "lazurio-secret-custody",
        "microsoft-copilot-architecture"
      ],
      "markdown": "<details>\n\n<summary>Is Lazurio an AI model?</summary>\n\nNo. Lazurio is the working environment and operating model around people,\nagents, repositories, modules and connected tools. A concrete installation\nuses a selected execution client and model provider whose terms must be\nreviewed separately.\n\n</details>\n\n<details>\n\n<summary>Does Lazurio replace Microsoft Copilot?</summary>\n\nNot categorically. Microsoft Copilot is a natural fit for assistance grounded\nin Microsoft 365. Lazurio focuses on governed, source-controlled work across\nrepositories and tools. Many organizations may use both. See the [full\ncomparison](/en/lazurio-vs-microsoft-copilot/).\n\n</details>\n\n<details>\n\n<summary>Can an agent access everything the user can?</summary>\n\nDo not assume that. Effective access depends on the session, tool, local\nworkspace and provider credential. The design principle is that a task agent\ndoes not gain rights from a prompt and should receive only the context needed\nfor the task. Your deployment must verify the actual grants and denied paths.\n\n</details>\n\n<details>\n\n<summary>Is all Lazurio data stored locally?</summary>\n\nNo universal claim is made. The documented root and working checkouts are\nlocal, while Git repositories, model requests, external applications and\ndeployed modules may use provider infrastructure. The concrete data-flow map\nbelongs to the deployment acceptance package.\n\n</details>\n\n<details>\n\n<summary>How are secrets handled?</summary>\n\nReal secrets stay outside Git in scoped, ignored custody paths or an approved\nprovider store. Tracked files contain only schemas, required variable names and\ninstructions. Read the public [secret custody standard](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/manual/security/local-secret-custody.md).\n\n</details>\n\n<details>\n\n<summary>What stops an agent from publishing a bad change?</summary>\n\nThe operating model distinguishes an editable draft from publication.\nRepository permissions, branch rules, required checks and reviews enforce Git\npublication; other systems need equivalent provider-side permissions and\nexplicit authorization. These controls reduce risk but do not make human\napproval infallible.\n\n</details>\n\n<details>\n\n<summary>Is Lazurio certified for a particular compliance framework?</summary>\n\nThis documentation makes no certification or universal compliance claim.\nOrganizations must evaluate their actual deployment, providers, controls and\nlegal obligations.\n\n</details>\n\n<details>\n\n<summary>Is there an audit trail?</summary>\n\nGit changes can be tied to commits, pull requests, reviews and deployments.\nProvider and endpoint actions require their own logs. Ask for an audit map that\nstates which system records each relevant action and how long it is retained.\n\n</details>\n\n<details>\n\n<summary>Does the documentation have an MCP server?</summary>\n\nNot yet. Agents can use [`llms.txt`](/llms.txt) and the structured\n[`content-index.json`](/content-index.json) today. A future MCP server is\nplanned as a read-only view over that same public content, not a second source.\n\n</details>\n\n<details>\n\n<summary>Where should I start an IT review?</summary>\n\nUse the [ten-minute IT briefing](/en/it-administrators/), then review [data\naccess and security](/en/data-access-security/) and [deployment and\noperations](/en/deployment-operations/) against the proposed configuration.\n\n</details>"
    },
    {
      "stableId": "lazurio-doc-how-it-works",
      "locale": "en",
      "route": "/en/how-lazurio-works/",
      "canonicalUrl": "https://documentation.lazurio.ai/en/how-lazurio-works/",
      "sourcePath": "data/v2/docs/en/how-lazurio-works.md",
      "title": "How Lazurio works",
      "summary": "Understand Lazurio's principals, agents, Organizations, repositories, workspace modules, drafts, reviews, and publication flow.",
      "audience": [
        "it-admin",
        "decision-maker",
        "builder",
        "agent"
      ],
      "updatedAt": "2026-08-26",
      "reviewedAt": "2026-08-26",
      "reviewOwner": "Matej Suchanek",
      "secondReviewOwner": null,
      "trustCritical": false,
      "sourceRefs": [
        "lazurio-readme",
        "lazurio-architecture",
        "lazurio-collaboration-model"
      ],
      "markdown": "Lazurio treats AI-assisted work as normal organizational work with explicit\nownership and boundaries. A person starts with a desired outcome; an agent can\ninspect the authorized context, create an editable draft, verify it and prepare\nit for review. Authority does not come from the agent's confidence. It comes\nfrom the signed-in identity and the systems that own the action.\n\nThe current public model is defined in the [Lazurio architecture](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/ARCHITECTURE.md)\nand [agent collaboration contract](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/AGENTS.md).\n\n## The five useful concepts\n\n**Principal** is the person or AI colleague for whom work is being done. The\nprincipal holds real permissions and the last word.\n\n**Task agent** is the active tool session. It works for the principal and has\nno independent authority. It can propose and prepare; a prompt does not make it\nan administrator.\n\n**Organization** is one company's access boundary. In the documented model it\nmaps to a GitHub organization and a separate repository root. Company-specific\ndata and strategy stay within that Organization.\n\n**Workspace module** is an application or bounded work area inside an\nOrganization. A module owns its runtime contract and can be developed,\nreviewed, deployed and rolled back independently.\n\n**Personalspace** is a private area for one principal. It is not an\norganizational collaboration store and is never a shortcut for moving company\ndata across access boundaries.\n\n## From request to published result\n\n1. **Scope:** identify the Organization, module and intended outcome.\n2. **Authority:** use the signed-in principal's live access; do not invent a\n   second permission system in prose.\n3. **Context:** load only the repositories and tools required for the task.\n4. **Draft:** make the work in a reversible form, commonly a Git branch and\n   pull request.\n5. **Evidence:** run checks, show the practical effect and preserve review\n   context.\n6. **Decision:** an authorized principal approves or rejects the exact result.\n7. **Publication:** merge, deploy, send or otherwise make the result effective.\n8. **Closeout:** update the authoritative plan, record remaining issues and\n   clean temporary workspaces.\n\nThis shape makes human review useful: the reviewer sees not only generated\ntext, but the exact change, evidence, owner and publication decision.\n\n## Source of truth, not one giant database\n\nLazurio does not require every kind of information to be copied into a single\nAI store. Code stays in repositories, plans stay in the Organization's Mission\nControl, durable knowledge stays in its Knowledgebase, and provider data stays\nbehind a scoped integration. The working environment brings the relevant\npieces together for a task while preserving their natural owners.\n\nThat distinction matters operationally. Removing repository access or\nrevoking an integration changes what the principal and its agent can reach;\nediting a role name in documentation does not.\n\n## Current state and target direction\n\nThe public architecture distinguishes a target model from currently deployed\nconfiguration. Not every installation exposes every planned capability. Ask\nfor live configuration and provider readback when evaluating a specific\ndeployment. The [IT briefing](/en/it-administrators/) lists the minimum\nevidence expected for approval."
    },
    {
      "stableId": "lazurio-doc-overview",
      "locale": "en",
      "route": "/en/",
      "canonicalUrl": "https://documentation.lazurio.ai/en/",
      "sourcePath": "data/v2/docs/en/index.mdx",
      "title": "Understand Lazurio before you approve it",
      "summary": "Start here to understand what Lazurio is, what problem it solves, and which trust and operational questions to ask.",
      "audience": [
        "it-admin",
        "decision-maker",
        "builder",
        "agent"
      ],
      "updatedAt": "2026-08-26",
      "reviewedAt": "2026-08-26",
      "reviewOwner": "Matej Suchanek",
      "secondReviewOwner": null,
      "trustCritical": false,
      "sourceRefs": [
        "lazurio-readme",
        "lazurio-architecture",
        "lazurio-collaboration-model"
      ],
      "markdown": "Lazurio is a working environment and operating model for collaboration between\npeople and AI agents. It gives that collaboration an explicit shape: companies\nare separated into Organizations, work lives in versioned repositories, access\ncomes from the systems that already own it, and consequential changes stay\nreviewable before publication. The public source describes both the target\n[architecture](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/ARCHITECTURE.md)\nand the current [collaboration rules](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/AGENTS.md).\n\nIt is not a promise that an AI system can safely do anything. It is a framework\nfor making authority, context, work and publication boundaries visible enough\nto govern.\n\n## Choose your path\n\n- **Deciding whether to allow it:** read the [ten-minute IT\n  briefing](/en/it-administrators/) and the [data-access\n  review](/en/data-access-security/).\n- **Understanding the mechanics:** see [how Lazurio\n  works](/en/how-lazurio-works/) and how it is [deployed and\n  operated](/en/deployment-operations/).\n- **Comparing alternatives:** use the sourced [Lazurio vs Microsoft\n  Copilot](/en/lazurio-vs-microsoft-copilot/) decision guide. They solve\n  overlapping, but not identical, problems.\n- **Letting an agent read the docs:** give it `/llms.txt` or\n  `/content-index.json`. Both are generated from the same reviewed pages that\n  people read."
    },
    {
      "stableId": "lazurio-doc-it-administrators",
      "locale": "en",
      "route": "/en/it-administrators/",
      "canonicalUrl": "https://documentation.lazurio.ai/en/it-administrators/",
      "sourcePath": "data/v2/docs/en/it-administrators.md",
      "title": "A ten-minute IT briefing",
      "summary": "A concise IT review of Lazurio's purpose, identity, access, data, integrations, operations, and approval boundaries.",
      "audience": [
        "it-admin",
        "decision-maker",
        "agent"
      ],
      "updatedAt": "2026-08-26",
      "reviewedAt": "2026-08-26",
      "reviewOwner": "Matej Suchanek",
      "secondReviewOwner": "Pablo AI",
      "trustCritical": true,
      "sourceRefs": [
        "lazurio-readme",
        "lazurio-architecture",
        "lazurio-collaboration-model",
        "lazurio-external-apps",
        "lazurio-secret-custody"
      ],
      "markdown": "Lazurio is a source-controlled working environment for people and AI agents.\nIt is designed to make organizational boundaries, authority and publication\ndecisions explicit while agents help with work. The public source is available\nfor inspection in the [Lazurio repository](https://github.com/HumanAndMachines/Lazurio/tree/69c53ec342124aef48cb9d04fd109f9886ec242e).\n\nThe right approval question is not “Can the AI see data?” in isolation. It is:\n**which identity is operating, on which machine, in which Organization, through\nwhich approved integration, against which data, and who may publish the result?**\n\n## The short version\n\n| Area | Documented Lazurio position | What IT should verify for its deployment |\n| --- | --- | --- |\n| Identity | The signed-in principal supplies authority; a task agent does not gain independent rights from a prompt. | The human or service identity, repository membership and device owner are correct. |\n| Organization boundary | One company maps to one Organization and GitHub organization/access boundary. Cross-organization data must not be mixed. | The intended repositories and teams are the only ones mounted and accessible. |\n| Local workspace | Work starts from checked-out, versioned source on a principal-owned machine. | Device hardening, disk encryption, endpoint monitoring, backup and offboarding meet policy. |\n| Publication | Agent work is a reversible draft. Merge, deployment, sending and other external publication require explicit authority. | Repository rules, required reviews and deployment permissions enforce the intended gate. |\n| External apps | Integrations are machine-local, reviewed and separately revocable; the documented preference is official MCP, then official CLI. | Each enabled provider, OAuth scope, data flow, retention policy and revocation path is accepted. |\n| Secrets | Secrets belong in ignored, scoped custody paths—not in Git or public documentation. | The chosen secret store, rotation, incident response and leak scanning are operational. |\n| Audit | Git commits, pull requests, reviews and provider logs create evidence, but coverage depends on the actual tools used. | Required logs exist across GitHub, endpoint, model provider, apps and deployment infrastructure. |\n\nThese are documented design and process boundaries, not a certification. The\nunderlying contracts are the public [collaboration model](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/AGENTS.md),\n[integration standard](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/manual/external-app-integrations.md),\nand [secret custody standard](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/manual/security/local-secret-custody.md).\n\n## What data can be reached?\n\nThere is no honest universal list because Lazurio is a framework whose concrete\nreach depends on the principal, machine, repository grants and enabled tools.\nA deployment review should inventory four surfaces:\n\n1. Git repositories and teams visible to the operating identity.\n2. Local files intentionally placed inside the active workspace boundary.\n3. External applications enabled through an approved MCP server or CLI.\n4. Model and hosting providers used by the chosen execution client and modules.\n\nAn integration existing in the ecosystem is not evidence that it is enabled.\nDemand a live, deployment-specific list with provider, scopes, owner and\nrevocation instructions.\n\n## What prevents accidental publication?\n\nThe [Lazurio collaboration contract](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/AGENTS.md)\nseparates editable agent work from publication. Normal development happens on\na branch and through a pull request. GitHub permissions and branch rules remain\nthe access authority; text in a prompt is not an access grant. For non-Git\nactions such as sending a message or changing a provider, the same principle is\nimplemented by explicit authorization and provider permissions.\n\nProcess controls still need technical counterparts wherever the platform can\nenforce them. Your acceptance test should try a denied repository, an\nunapproved external tool, a protected merge and a revoked credential—not only\nread the policy.\n\n## Minimum approval package\n\nBefore a production rollout, ask the operator for:\n\n- the Organization and repository inventory;\n- named human/service identities and GitHub team grants;\n- device and local data protection baseline;\n- model provider and data-processing terms for the selected client;\n- integration catalogue with exact scopes and revocation owners;\n- secret custody, backup, deletion and offboarding procedures;\n- branch protection and publication authority;\n- logging sources, retention and incident contact;\n- a bounded pilot with data that represents the real use case.\n\nIf any answer is unknown, record it as an implementation issue rather than\nturning an architectural intention into a security claim.\n\n## Suggested decision\n\nApprove a scoped pilot when the identities, repositories, integrations, model\nprovider and publication gate are concrete and testable. Do not approve a\nblanket deployment from this overview alone. Start with [data access and\nsecurity](/en/data-access-security/) and [deployment and\noperations](/en/deployment-operations/) for the deeper review."
    },
    {
      "stableId": "lazurio-doc-copilot-comparison",
      "locale": "en",
      "route": "/en/lazurio-vs-microsoft-copilot/",
      "canonicalUrl": "https://documentation.lazurio.ai/en/lazurio-vs-microsoft-copilot/",
      "sourcePath": "data/v2/docs/en/lazurio-vs-microsoft-copilot.md",
      "title": "Lazurio vs Microsoft Copilot",
      "summary": "Compare Lazurio with Microsoft Copilot across purpose, context, permissions, execution, extensibility, governance, and deployment.",
      "audience": [
        "it-admin",
        "decision-maker",
        "agent"
      ],
      "updatedAt": "2026-08-26",
      "reviewedAt": "2026-08-26",
      "reviewOwner": "Matej Suchanek",
      "secondReviewOwner": "Pablo AI",
      "trustCritical": true,
      "sourceRefs": [
        "lazurio-readme",
        "lazurio-architecture",
        "lazurio-collaboration-model",
        "lazurio-external-apps",
        "microsoft-copilot-architecture",
        "microsoft-copilot-data-protection",
        "microsoft-copilot-privacy",
        "microsoft-copilot-requirements",
        "microsoft-copilot-extensibility"
      ],
      "markdown": "Lazurio and Microsoft Copilot overlap in one important way: both help people\nuse AI in real work. They are not direct substitutes at every layer.\n\n**Microsoft Copilot** is Microsoft's AI experience embedded across Microsoft\n365 and grounded in the signed-in user's permitted work data. **Lazurio** is an\nopen, source-controlled operating model and workspace for governed work by\npeople and agents across repositories, modules and explicitly connected tools.\n\nThis page uses “Microsoft Copilot” for the current product family while the\nlinked provider pages may use “Microsoft 365 Copilot.” Microsoft product facts\nbelow come from current Microsoft Learn documentation. Lazurio positioning and\nthe recommendations are our assessment.\n\n## Side-by-side\n\n| Decision area | Microsoft Copilot | Lazurio |\n| --- | --- | --- |\n| Primary purpose | AI assistance inside the Microsoft 365 productivity environment. | Governed execution and durable collaboration across source-controlled organizational work. |\n| Natural context | Microsoft Graph and the user's permitted Microsoft 365 content, plus configured agents and connectors. | Selected Organization repositories, local workspace context and individually approved external tools. |\n| Permission foundation | The signed-in user's Microsoft 365 permissions, identity controls and service boundary. | The signed-in principal's GitHub/provider permissions and machine-scoped integrations; prompts do not grant authority. |\n| Typical output | Answers, summaries, drafts and actions in Microsoft applications and Copilot experiences. | Reviewable changes, plans, knowledge, applications and tool actions prepared through a defined draft-to-publication flow. |\n| Control plane | Microsoft-administered tenant services, licensing, Purview, Entra and Microsoft 365 administration. | Organization-owned Git repositories, machines, module contracts and each connected provider's own administration. |\n| Extensibility | Microsoft agents, Graph connectors, Copilot APIs and SDKs. | Modules, agent competence packages, MCP servers, official CLIs and browser workflows under local curation. |\n| Deployment | Microsoft-managed cloud service with published tenant requirements. | Framework and module deployment chosen per Organization; current public model begins with local checkouts and independently deployed modules. |\n| Best fit | Productivity and knowledge work already centered on Microsoft 365. | Work that must cross repositories or tools while remaining source-controlled, testable and explicitly publishable. |\n\nMicrosoft documents that Copilot works within the Microsoft 365 service\nboundary, uses Microsoft Graph and only presents data a user is authorized to\naccess. See Microsoft's [architecture overview](https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-architecture)\nand [data protection and auditing](https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-architecture-data-protection-auditing).\n\nLazurio's public source defines GitHub as the access authority for repository\nwork, separates Organizations, and treats agent output as a draft until an\nauthorized publication decision. See the [collaboration contract](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/AGENTS.md)\nand [architecture](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/ARCHITECTURE.md).\n\n## Data and privacy questions\n\nMicrosoft states that prompts, responses and Microsoft Graph data used by\nMicrosoft 365 Copilot are not used to train its foundation models. It also\ndocuments stored interaction history, Purview controls and the need to assess\nterms for third-party agents, connectors or model providers. Read the current\n[Microsoft privacy and security statement](https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-privacy)\nfor the exact product and tenant terms.\n\nLazurio cannot give one equivalent model-provider statement because it is not\none universal AI service account. Its data path depends on the selected\nexecution client, model provider, Organization grants and integrations. A\nLazurio approval therefore needs a deployment-specific provider and scope\ninventory. This flexibility is valuable when an organization needs a tailored\ntoolchain, but it creates more configuration responsibility.\n\n## Administration and prerequisites\n\nMicrosoft publishes tenant prerequisites including eligible licensing,\nMicrosoft Entra ID accounts, supported update channels and network endpoints;\nSharePoint and Microsoft Purview are part of its recommended preparation. See\nthe current [Microsoft requirements](https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-minimum-requirements).\n\nLazurio instead requires the organization to own its GitHub structure,\nendpoint baseline, execution provider, repository rules and enabled\nintegrations. It offers more direct control of the working source and\npublication process, while leaving more operational choices to the\norganization.\n\n## Extensibility is strong in both, but different\n\nMicrosoft Copilot can be extended through agents, Microsoft Graph connectors,\nCopilot APIs and SDKs documented in its [extensibility overview](https://learn.microsoft.com/en-us/microsoft-365/copilot/extensibility/).\nThat is the natural route when the workflow should live inside Microsoft's\nproduct and governance ecosystem.\n\nLazurio extensions are workspace modules and scoped tool integrations. The\ndocumented [external application standard](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/manual/external-app-integrations.md)\nprefers official local MCP servers or CLIs and separately revocable machine\nidentities. That is useful when the workflow is source-centric, crosses\ndifferent providers or needs custom verification and publication gates.\n\n## Which should you choose?\n\n### Choose Microsoft Copilot first when\n\n- most valuable work already happens in Outlook, Teams, Word, Excel,\n  PowerPoint and SharePoint;\n- the tenant's Microsoft 365 permission hygiene and governance are mature;\n- you want a vendor-managed productivity experience with Microsoft's admin\n  and compliance surfaces;\n- custom work can stay within Microsoft's agent and connector ecosystem.\n\n### Choose Lazurio first when\n\n- the output must become a reviewed repository change, module, operating plan\n  or durable organizational source of truth;\n- work crosses GitHub, local source and several non-Microsoft providers;\n- you need explicit company boundaries and an inspectable draft-to-publication\n  workflow;\n- your organization is prepared to own endpoint, repository, provider and\n  integration configuration.\n\n### Use both when\n\nMicrosoft Copilot is the productivity assistant for Microsoft 365 while\nLazurio governs source-controlled delivery and cross-tool execution. Keep the\nresponsibilities and credentials distinct: do not assume an approval or data\nboundary in one automatically applies to the other.\n\n## Bottom line\n\nMicrosoft Copilot is usually the simpler answer for Microsoft 365-centered\npersonal and team productivity. Lazurio is aimed at a broader operational\nproblem: turning agent-assisted work across repositories and tools into\nreviewable, owned and publishable organizational outcomes. A pilot should test\nthe real workflow, the permission model and the operational burden—not just\ncompare chat quality."
    },
    {
      "stableId": "lazurio-doc-reference",
      "locale": "en",
      "route": "/en/reference/",
      "canonicalUrl": "https://documentation.lazurio.ai/en/reference/",
      "sourcePath": "data/v2/docs/en/reference.md",
      "title": "Public references",
      "summary": "Inspect the pinned Lazurio source and current Microsoft Learn pages behind architectural, security, privacy, and comparison claims.",
      "audience": [
        "it-admin",
        "decision-maker",
        "builder",
        "agent"
      ],
      "updatedAt": "2026-08-26",
      "reviewedAt": "2026-08-26",
      "reviewOwner": "Matej Suchanek",
      "secondReviewOwner": "Pablo AI",
      "trustCritical": true,
      "sourceRefs": [
        "lazurio-readme",
        "lazurio-architecture",
        "lazurio-collaboration-model",
        "lazurio-external-apps",
        "lazurio-secret-custody",
        "microsoft-copilot-architecture",
        "microsoft-copilot-data-protection",
        "microsoft-copilot-privacy",
        "microsoft-copilot-requirements",
        "microsoft-copilot-extensibility"
      ],
      "markdown": "The first release uses public, inspectable sources. Lazurio links are pinned to\nthe exact source revision reviewed for this documentation. Microsoft links\npoint to current provider documentation because the service changes over time;\ntheir shorter review window is enforced by the source map.\n\n## Lazurio source\n\n- [Project overview](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/README.md)\n- [Architecture](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/ARCHITECTURE.md)\n- [Collaboration, authority and publication model](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/AGENTS.md)\n- [External application integration standard](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/manual/external-app-integrations.md)\n- [Local secret custody standard](https://github.com/HumanAndMachines/Lazurio/blob/69c53ec342124aef48cb9d04fd109f9886ec242e/manual/security/local-secret-custody.md)\n\n## Microsoft provider documentation\n\n- [Microsoft 365 Copilot architecture](https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-architecture)\n- [Data protection and auditing](https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-architecture-data-protection-auditing)\n- [Data, privacy, and security](https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-privacy)\n- [Requirements](https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-minimum-requirements)\n- [Extensibility](https://learn.microsoft.com/en-us/microsoft-365/copilot/extensibility/)\n\n## Review contract\n\nMachine-readable source metadata lives in the public repository at\n`data/v2/source-map.json`. Each entry names an owner, review date and next\nreview deadline. Content validation fails after a source expires, preventing a\nsilent claim that old provider behavior is still current.\n\nSecurity-critical pages also name Pablo AI as second reviewer. Production\npublication requires approval of the exact Git commit, not a general approval\nof an earlier draft."
    },
    {
      "stableId": "lazurio-doc-use-cases",
      "locale": "en",
      "route": "/en/use-cases/",
      "canonicalUrl": "https://documentation.lazurio.ai/en/use-cases/",
      "sourcePath": "data/v2/docs/en/use-cases.md",
      "title": "What Lazurio is good for",
      "summary": "See where Lazurio helps with product delivery, operations, knowledge, agent workflows, and multi-organization work—and where it does not.",
      "audience": [
        "it-admin",
        "decision-maker",
        "builder",
        "agent"
      ],
      "updatedAt": "2026-08-26",
      "reviewedAt": "2026-08-26",
      "reviewOwner": "Matej Suchanek",
      "secondReviewOwner": null,
      "trustCritical": false,
      "sourceRefs": [
        "lazurio-readme",
        "lazurio-architecture",
        "lazurio-collaboration-model"
      ],
      "markdown": "Lazurio is most useful when AI assistance needs to become repeatable,\nreviewable organizational work rather than a sequence of private chat answers.\nIts public model combines separate company boundaries, versioned sources of\ntruth, tool access and explicit publication decisions.\n\n## Strong-fit use cases\n\n### Product and software delivery\n\nAn agent can inspect a module, implement a bounded change, run its tests and\nprepare a pull request with evidence. The team keeps normal Git ownership,\nreview and rollback instead of accepting an opaque generated artifact.\n\n### Operational workflows across tools\n\nA task may need repository context plus an approved external application. The\nLazurio integration model makes each provider connection an explicit,\nrevocable part of the machine rather than assuming every agent should have a\nuniversal cloud broker.\n\n### Durable organizational knowledge\n\nDecisions, plans, issues and reusable knowledge can be written back to the\nOrganization's named source of truth. The chat remains a working surface, not\nthe only place where context lives.\n\n### Repeated agent roles\n\nOrganizations can package scoped instructions, tools, tests and publishing\ngates for recurring work. That makes the next run start from a reviewed\ncompetence boundary rather than a copied prompt with hidden assumptions.\n\n### Multiple companies on one principal's machine\n\nThe root can discover several authorized Organizations while keeping each one\na separate GitHub and repository boundary. This is useful for founders,\nconsultants and operators who genuinely work across companies and must not mix\ntheir private data.\n\n## Use cases that need additional controls\n\nRegulated data, production infrastructure, financial actions, bulk messaging,\ncredential administration and destructive operations can be supported only\nwhen the concrete provider, access and approval controls are designed for that\nrisk. Lazurio's process model is not a substitute for applicable law,\ncertification, segregation of duties or provider security review.\n\n## When Lazurio may not be the best first choice\n\n- You only need personal drafting inside one productivity suite and do not need\n  source-controlled execution or cross-tool workflows.\n- The organization is unwilling to own repository access, endpoint controls or\n  review rules.\n- The desired task cannot be bounded, tested or reversed and there is no\n  authorized human decision point.\n- A mature product already solves the whole workflow with a simpler supported\n  control plane.\n\nFor Microsoft-centric personal productivity, read the [Lazurio vs Microsoft\nCopilot](/en/lazurio-vs-microsoft-copilot/) guide. The sensible result may be\nCopilot, Lazurio, or both with separate responsibilities."
    }
  ]
}