CD Consulting R&D

KB-AI · Entry 2 · Published 2026-10-11 · Version 1

One model, many doors: a dated map of every way to reach Claude in October 2026, and what actually differs between them

A dated map of every door to Claude, from the chat app and the Office add-ins to Claude Code, the developer platform and the three cloud marketplaces: where each one runs, who pays, what persists, and what you give up by choosing one over another.

If you have ever wondered why "Claude" is sold as an app, a developer tool, a cloud session and an Amazon product at the same time, this is the one-page map of the doors, with where each runs, who pays and what differs, as of October 2026.

Analysis from Anthropic's documentation, help centre and announcements consulted on 11 October 2026, and from two user conversations of 8 and 10 October 2026. List prices in US dollars. The state described is the state on that date.

Contents

1. 1. The question behind the confusion
2. 2. The map in one table
3. 3. The Claude apps: one Claude since 16 September 2026
4. 4. Claude Code: one engine, six surfaces, and five ways to be away from it
5. 5. Building on Claude: the platform and the three resellers
6. 6. Where the context lives: a folder or an account
7. 7. Who pays, and what the credit was for
8. 8. What happens to the data
9. 9. How to choose, and the limits of this map
10. 10. Sources

1. The question behind the confusion

Anyone who has used Claude for more than a few weeks ends up asking a version of the same question. One user put it to Claude directly on 8 October 2026: what fundamentally distinguishes a chat session from a Claude Code session, when both can read the same project instructions, use the same skills and open the same GitHub repository? Two days later, having received a notice of a credit "for cloud sessions", the same user asked what that credit could possibly pay for if a chat already runs in the cloud and can be pointed at a repository. The answers that came back are the seed of this entry [S27], and they are worth restating because they hold the key to the whole map.

The first answer: on the inside there is almost no difference. The chat session and the coding tool run the same agent engine, with the same file tools, the same shell, the same sub-agents, the same mechanism for loading a CLAUDE.md file of project instructions. What differs is the environment that engine is launched into: a folder on your machine, or an ephemeral container in Anthropic's cloud. Anthropic's own documentation now says this in as many words: "Claude Code runs the same underlying engine everywhere, but each surface is tuned for a different way of working" [S1].

The second answer: separate two questions that the marketing runs together. Where does the session run? And who triggers it and who pays for it? Once those two are held apart, every product name on Anthropic's price list falls into place.

This entry applies that lens to the whole catalogue as it stood on 11 October 2026. It is deliberately dated. In the four weeks before that date Anthropic merged two of its consumer products into one [S21], launched a credit promotion that expires on 4 November [S23], and kept at least one documentation page that still describes the pre-merger layout [S7]. A map of this territory has a short shelf life; the method of reading it does not.

2. The map in one table

Four families cover every door. The first is the Claude apps: the consumer and team product, reached through a browser, a desktop application, a phone, a browser extension or an Office add-in, and paid for by subscription. The second is Claude Code: the agent for software and, increasingly, for any file-based work, reached through a terminal, an editor, the desktop app, the browser, a phone or Slack, paid for by the same subscription or by API consumption. The third is the Claude Developer Platform: the API and the hosted agent service, for people who write code against the model, paid per token. The fourth is the cloud marketplaces: the same models resold inside Amazon Web Services, Google Cloud and Microsoft Azure, invoiced by those providers.

Three things stand out before any detail. First, the row that runs on your own machine is a minority: most doors run somewhere Anthropic or a cloud provider controls. Second, there are only two ways to pay, a subscription or consumption, and every door uses exactly one of them except the Max plan, which bundles some API credit into a subscription [S17]. Third, what persists is never the same object: an account, a folder, a branch, a session store or an audit log. That third column is where most surprises come from.

3. The Claude apps: one Claude since 16 September 2026

Until mid-September 2026 the consumer product had two modes. "Chat" was the conversation. "Cowork", launched as a research preview in January 2026, was a desktop agent that could work in your files, run for a long time and report back. On 16 September Anthropic announced that the two are "now one Claude": the user no longer picks a mode, and Claude "determines what a task needs" [S21]. The rollout started with Pro and Max subscribers on web, desktop and mobile, with Team and Free "soon" and Enterprise organisations given at least thirty days' notice [S21] [S22]. On 11 October the help centre still described the rollout as gradual, with accounts on the same plan receiving it at different times, and listed what the unified experience could not yet do: add a repository from GitHub, branch a conversation from an earlier point, search older Cowork tasks, or offer Dispatch to new users [S22]. Once an account has moved over, it cannot go back [S22].

What the merger means in practice is that a single conversation can now produce a document, a slide deck, a design, a spreadsheet or a file on disk, and can keep working in the cloud after the laptop is closed [S21]. The desktop app adds what only a desktop can: access to granted folders, a built-in browser and, in beta on Pro and Max, "computer use", where Claude operates applications by clicking and typing [S22]. The mobile app shows the same tasks, lets you answer questions and redirect work, and can start tasks that then continue in the cloud [S22].

Two companions extend the app into places it did not reach. Claude in Chrome is a browser extension that reads, clicks and navigates alongside you; on Max and Team plans, and rolling out to Pro, its side panel is itself a Cowork-style session whose history syncs to web, desktop and mobile, and it can save shortcuts and run browser tasks on a daily, weekly or monthly schedule [S10]. Anthropic's own help page says the extension "is still risky" and that it can intervene when it detects a high-risk website [S10]. The same extension is what Claude Code drives from the terminal or VS Code to test a web application with the user's logged-in sessions; that path needs a claude.ai sign-in and is off when the tool is authenticated with an API key or a cloud provider [S9]. Claude for Microsoft 365 is a set of add-ins for Excel, PowerPoint, Word and Outlook that share one conversation across the four applications, so a change made in a spreadsheet is known to the deck being drafted next door [S20]. Enterprises can route the add-ins through Amazon Bedrock, Google Cloud Vertex AI, Azure AI Foundry or an LLM gateway rather than through Anthropic directly [S20].

The pricing page, consulted the same day, lists the plans as follows. Free: chat on web, desktop and mobile, web search, file creation and code execution, memory, artifacts, five projects, Sonnet and Haiku only, no Claude Code. Pro, at $20 a month or $17 billed annually: everything in Free plus Claude Code, Claude in Chrome, Claude for Microsoft 365, scheduling, unlimited projects and Opus; the most capable model, Fable, is available through purchased usage credits. Max, from $100 a month: five or twenty times Pro's usage, Fable at half the weekly limit, and $100 or $200 a month of API credit included. Team: $20 or $25 per standard seat, $100 or $125 per premium seat, with pooled API credit up to $500 a month. Enterprise: $20 per seat billed annually plus usage at API rates, with spend limits, role-based access, SCIM, audit logs, a compliance API, custom data retention and IP allow-listing [S17]. The structure to notice is that the apps and Claude Code are sold together: a Pro seat is a seat for both.

4. Claude Code: one engine, six surfaces, and five ways to be away from it

Claude Code began as a terminal tool. By October 2026 Anthropic lists six surfaces for it, all running "the same underlying engine" [S1]:

• The CLI, in a terminal, which remains the most complete surface: scripting, the Agent SDK and third-party providers are CLI-only or CLI-first [S1].
• Desktop, the "Code" tab of the Claude desktop application on macOS, Windows and Linux, which adds a diff viewer, app previews, parallel sessions in isolated git worktrees, an integrated terminal and connectors [S7].
• VS Code and JetBrains extensions, which run the engine inside the editor with inline diffs and selection sharing [S1].
• The web, at claude.ai/code, where sessions run in the cloud and "continue after you disconnect" [S1] [S4].
• Mobile, where the Claude app for iOS and Android is "a client for Claude Code sessions rather than a place where code runs" [S6].

The decisive line runs between the local surfaces and the cloud ones. A local session is anchored in a folder: it reads that folder's CLAUDE.md and .claude/ directory, uses the tools installed on the machine, and commits to the local repository. Configuration, project memory and MCP servers are shared across the local surfaces [S1]. A cloud session is anchored in a GitHub repository: an Anthropic-managed virtual machine clones the repository, works on a branch, and pushes it back through a proxy that holds the GitHub credential on the server side so that the credential never enters the VM [S4]. The session counts against the plan's usage limits "alongside the rest of your Claude and Claude Code usage, and there's no separate charge for the cloud VM" [S4]. Cloud sessions are available on Pro, Max and Team plans, and on Enterprise with a premium or "Chat + Claude Code" seat; they require a claude.ai sign-in and are not reachable with an API key or through Bedrock, Vertex or Foundry [S2] [S4]. Organisations that want the same thing inside their own network can deploy self-hosted environments, where the cloud session runs on runners they operate [S1] [S4].

The two are joined by two one-way commands. claude --cloud "task" starts a new cloud session from a terminal, cloning the repository's GitHub remote at the current branch, "not your local checkout" [S4]. claude --teleport pulls a finished cloud session, its branch and its conversation back into the terminal, where it becomes a separate copy [S4]. A repository without a GitHub remote can be bundled and uploaded instead, with uncommitted changes to credential-shaped files left out on macOS and Linux but, the documentation warns, uploaded as they are on native Windows [S4].

Around these surfaces Anthropic has built five ways to work "when you are away from your terminal", which it compares in a single table by trigger, by where Claude runs and by setup [S1]:

• Dispatch: message a task from the phone; it runs on your own machine through the desktop app, which must stay open and awake [S1] [S24]. Dispatch is in limited beta on some Pro and Max accounts and is not open to new users since the merger [S22] [S24].
• Remote Control: steer a session that is already running on your machine from the browser or the phone; execution and file access stay local, while the transcript is also stored on Anthropic's servers to sync devices [S1] [S11].
• Channels: push events from Telegram, Discord or your own server into a local session, through a plugin [S1].
• Slack: mention @Claude in a channel under your own account, on Pro and Max, and the work runs in Anthropic's cloud [S1]. For organisations, Claude Tag does the same under a shared identity configured by an admin, in an ephemeral sandbox, billed by usage from an organisation balance rather than per seat, on Team and Enterprise only [S19].
• Routines: a saved prompt, repositories and connectors packaged once and run on a schedule, on an authenticated HTTP call, or on a GitHub pull-request or release event; each run is a full cloud session with no permission prompts, draws on subscription usage, and is in research preview [S5]. A local counterpart, Desktop scheduled tasks, runs on your machine instead [S1] [S5].

A sixth piece, Projects in the Code sense, is one conversation in which Claude starts and tracks many parallel cloud sessions for a body of work; it is in public beta on Pro and Max, rolling out first to accounts that have used cloud sessions and do not already have Projects in chat or Cowork, and not yet on Team or Enterprise [S8]. The name collides with the older "Projects" of the chat app, which are folders of instructions and files; the collision is itself a sign of how fast the catalogue is moving.

5. Building on Claude: the platform and the three resellers

The Claude Developer Platform offers "two ways to build" [S13]. The Messages API is direct access to the model: your code sends a conversation and receives a reply, and if you want an agent you write the tool loop yourself, or let the client SDK's beta tool runner drive it [S12]. Claude Managed Agents is the opposite bet: Anthropic runs the agent loop, the sandbox, prompt caching and context compaction in managed infrastructure; you define an agent (model, system prompt, tools, MCP servers, skills), an environment (Anthropic's cloud sandbox or a self-hosted one), start a session and exchange events with it over server-sent events [S14]. It is in beta, requires a dated beta header on every request, supports scheduled deployments on a cron, and, because it is stateful by design, is "not currently eligible" for zero data retention or HIPAA coverage [S14]. It is billed on tokens plus $0.08 per session-hour of running time [S16].

Between the two sits the Agent SDK, which packages the Claude Code engine itself as a Python or TypeScript library: the same built-in tools, hooks, sub-agents, permissions, sessions, skills and CLAUDE.md loading, "in a process you operate" [S12]. Anthropic's own comparison table places the four options by who runs the agent and what comes built in: the Agent SDK for embedding in your own application, the CLI for interactive use, the client SDK for calling the API directly, Managed Agents for having Anthropic host it [S12]. One restriction matters for anyone productising: Anthropic "does not allow third party developers to offer claude.ai login or rate limits for their products", so an agent built on the SDK must authenticate with an API key or a cloud provider, not with a user's subscription [S12].

The same four current models, Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, are listed with identifiers for five places: the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS [S15]. Anthropic's enterprise deployment page compares the options in one table: the Console is for individual developers and bills pay-as-you-go; Bedrock, the Agent Platform and Foundry are for organisations already on AWS, GCP or Azure, bill through those providers, authenticate with the provider's identity system and expose the provider's audit tooling (CloudTrail, Cloud Audit Logs, Azure Monitor); and only a Team or Enterprise subscription "includes Claude on web" [S3]. Two hybrids blur the line. Claude Platform on AWS is Anthropic's own API bought through AWS Marketplace: token usage is rated in dollars at the standard rates, converted to "Claude Consumption Units" at one cent each and metered hourly to AWS, which invoices monthly in arrears [S16]. Microsoft Foundry offers each model in two hosting options, "Hosted on Azure", where inference runs on Anthropic's service inside Azure and "prompts and completions remain within Azure", and "Hosted on Anthropic", where requests go to Anthropic's infrastructure; the most capable models, including Fable 5.1, are available only in the second [S18].

What an organisation gives up by buying through a reseller is spelled out on a single documentation page, which is the most useful page in this whole map for a buyer [S2]. Everything that runs locally in Claude Code works on every provider: the CLI, the editors, sub-agents, hooks, skills, CLAUDE.md, MCP servers, sandboxing. Everything that lives in Anthropic's cloud requires a claude.ai subscription and is absent on Bedrock, the Agent Platform, Foundry and with a bare API key: cloud sessions and mobile, Slack, the desktop app's Code tab (except through a separately documented enterprise build), Routines, Remote Control, the Chrome extension, computer use, artifacts and voice [S2]. Some server-side capabilities vary by provider: web search is absent on Bedrock, present on the Agent Platform only for Claude 4 models and later; fast mode and the Advisor exist only on Anthropic's own endpoints; GitHub Actions work everywhere except Claude Platform on AWS, GitLab CI/CD everywhere except Foundry [S2]. On the API side Foundry lacks the Admin API, Message Batches, the Models API and Managed Agents, and its Azure-hosted deployments additionally lack code execution, Agent Skills, programmatic tool calling and the Files API; requests that use them "return a 400 Bad Request error by design" [S18]. The pattern is consistent: the first-party endpoint receives every feature first, and the resellers carry a subset that grows over time.

A note on names, since they are a source of confusion in their own right. Google renamed Vertex AI to "Google Cloud's Agent Platform" during 2026; Anthropic's pages use the new name with "formerly Vertex AI" in brackets, while the SDK class, the environment variable and the endpoint host keep the old one [S2] [S3]. Microsoft's offering is "Microsoft Foundry" on Anthropic's pages and "Azure AI Foundry" on others [S20]. The model is the same; the catalogue pages are not.

6. Where the context lives: a folder or an account

The most useful distinction from the October conversations is not about features at all. It is about where persistent context lives, and it explains why the chat and the coding tool feel different even though they share an engine.

A local Claude Code session is anchored in a folder. The folder carries the project's standing instructions in CLAUDE.md, its settings and skills under .claude/, its git history and its installed tools. Change folder and you change context. Anthropic's enterprise guidance builds on exactly this: deploy CLAUDE.md files "at multiple levels", check an .mcp.json into the repository so that every developer inherits the same integrations, and pin model versions when buying through a cloud provider so that aliases do not drift [S3].

A chat session, and a cloud session, are anchored in an account. The user's memory, preferences, projects, account-level skills and connectors to Gmail, Drive, Dropbox or Microsoft 365 load at the start of every conversation without being asked for; what does not survive is the working disk, an ephemeral Linux container that is discarded when the session ends. A CLAUDE.md is loaded there only when a repository or a folder is attached, because the mechanism needs a starting directory. The user who tested this on 8 October found that moving an existing conversation into a project did not reload the project's instructions, while starting a fresh conversation inside the project did: project context, like CLAUDE.md, is read at start-up, not on the move.

This yields a rule that the documentation never states as plainly as the conversation did. Put what concerns a project in a CLAUDE.md at the root of its repository, and it will apply identically whether the repository is opened in a terminal, attached to a cloud session, cloned by a Routine or handed to an Agent SDK process. Put what concerns you, whatever the project, in account memory or in a skill, and it will follow you across the chat, the desktop app, the phone and the browser extension. The two mechanisms are complementary, not competing, and the "no watertight boundary" between chat and code that the user noticed is real: capabilities are shared, and only the default anchor differs.

The same lens explains the recurring complaint about prompt suggestions in the interface. They are generated outside the model's view, by the application; a suggestion never executes itself, but a reflexive keypress on an empty input can send one into a long-running session. Claude Code exposes a setting to turn them off, with a history of bugs around its persistence; the chat application had, as of October, no documented switch. The protection is the same as for any other interruption: permission prompts, plan mode, frequent commits.

7. Who pays, and what the credit was for

The cloud-session credit that puzzled the 10 October user is a clean illustration of the second question, who triggers and who pays. Anthropic's help centre explains it: a one-time bonus of $100 for Pro and $250 for Max subscribers, claimable until 7 October 2026 and expiring on 4 November, that "covers cloud sessions only" [S23]. It does not cover Projects, Routines, Remote Control sessions (which run locally), chat and Cowork sessions, or any session authenticated with an API key or through Bedrock or Vertex [S23]. While the credit lasts, cloud-session usage draws on it and does not count toward plan limits; afterwards, cloud sessions "count toward plan limits, just like local sessions", with no separate charge for the container [S23]. So the credit was not a new product. It was a subsidy for one trigger path, the cloud VM, inside a subscription that already paid for everything else.

The general shape of payment follows from the table in section 2. A subscription buys a bundle of surfaces (apps, Claude Code local and cloud, Chrome, Office, Routines) under rolling usage windows, with optional metered overage when "usage credits" are switched on [S17] [S23]. Consumption buys tokens: on the Claude API at list prices per million tokens, with batch processing at half price, cache reads at a fraction of input, and a 1.1 multiplier for US-only inference on models from the 4.6 generation onward [S16] [S17]. On the resellers, consumption is invoiced by the cloud provider, with regional and multi-region endpoints carrying a 10 % premium over global endpoints [S16]. Managed Agents adds a time dimension, $0.08 per session-hour, in place of the container-hour billing of the code-execution tool [S16]. Claude Tag adds a third model, an organisation balance drawn by usage under a spend limit, with no per-seat charge for the Slack users themselves [S19].

For a small practice or a single consultant, the practical consequence is that the subscription is the cheap door for interactive and scheduled work, and the API key is the right door for anything that must run unattended inside one's own pipeline, from a cron job or a GitHub Action, because that path is explicitly excluded from the subscription's cloud credit and from its surfaces [S23] [S12]. The two conversations reached the same split: use the chat and cloud sessions to design and test a procedure; use an API key and the Agent SDK or claude -p to run it in production.

8. What happens to the data

The door also decides the data regime, and here the line runs between consumer and commercial terms rather than between surfaces. On the consumer plans (Free, Pro, Max) Anthropic asks each user to choose whether conversations may be used to improve future models; the choice covers Claude Code sessions from those accounts, and retention follows it, five years when allowed and thirty days when not [S11] [S25]. The deadline for existing users to make the choice was 8 October 2025; the setting can be changed at any time in the account's privacy controls, and a deleted conversation is not used for future training [S25] [S26]. Under commercial terms (Team, Enterprise, the API and the three cloud providers) Anthropic "does not train generative models using code or prompts" unless the customer opts into a partner programme, retention is thirty days by default, and zero data retention is available to qualified Enterprise and API accounts on a per-organisation basis, not as part of the standard plan [S11] [S2].

The reseller route changes where data sits and what reaches Anthropic. On Bedrock and the Agent Platform, Claude Code's telemetry, error reporting and feedback are off by default, and encryption at rest is the provider's, with customer-managed keys available [S11]. On Foundry's Azure-hosted deployments, "prompts and completions remain within Azure; only usage metadata and content flagged by Anthropic's safety systems egress to Anthropic" [S11] [S18]. Cloud sessions, by contrast, store the transcript on Anthropic's servers so that the session can be resumed from any device, and route all outbound traffic through a security proxy [S11]. Remote Control keeps execution local but also stores the transcript on Anthropic's servers while connected [S11]. Managed Agents, being stateful, is outside zero-data-retention scope altogether [S14].

For a European organisation choosing among these doors, the decisive variables are therefore not model quality, which is identical, but residency and audit: the provider's regions and the 10 % premium for pinning them [S16], the provider's identity and logging stack [S3], and whether the Anthropic features that require a first-party endpoint (web search, Managed Agents, cloud sessions) are worth the first-party data path.

9. How to choose, and the limits of this map

A short decision guide follows from the four families, in the order a careful colleague would ask the questions.

1. Does the work need your machine? Local files, installed tools, a corporate network, a browser you are already logged into: a local Claude Code surface (CLI, editor, desktop), with Remote Control or Dispatch if you want to steer it from elsewhere [S1]. If it needs an Office file open, the Microsoft 365 add-in; if it needs a web application without an API, the Chrome extension [S10] [S20].
2. Does the work need to survive your laptop closing? A cloud session, a Routine, Claude Tag or, for application builders, Managed Agents [S4] [S5] [S14] [S19]. All four require Anthropic's first-party endpoint; none exists on a reseller [S2].
3. Who triggers it? A person in conversation: the app or a cloud session. A schedule or an event: a Routine (subscription) or a scheduled Managed Agents deployment (API). Your own code or pipeline: the Messages API, the Agent SDK or claude -p with an API key [S5] [S12] [S14].
4. Who must be invoiced, and where must the data stay? Your existing cloud bill and regions: Bedrock, the Agent Platform or Foundry, accepting the feature subset [S2] [S3]. Anthropic's bill with marketplace procurement: Claude Platform on AWS [S16]. Otherwise, a subscription for people and an API key for machines.

Limits. This entry rests on Anthropic's documentation and help centre as consulted on one day, 11 October 2026, and on two user conversations with Claude from the same fortnight; third-party coverage of the September merger was used only to locate the primary pages. Prices are list prices in US dollars and omit taxes, enterprise negotiation and the resellers' own tariff pages. Several features described are in beta or research preview (Managed Agents, Routines, Projects, Dispatch, computer use) and one documentation page still describes the three-tab desktop layout that the merger is replacing [S7], so the map should be re-read against the live pages before any purchasing decision. Nothing here evaluates model quality or compares Anthropic's catalogue with other vendors'; the subject is the doors, not what is behind them.

10. Sources

1. S1 Official (vendor documentation). Anthropic, Platforms and integrations, Claude Code documentation. https://code.claude.com/docs/en/platforms — consulted 11 October 2026.
2. S2 Official (vendor documentation). Anthropic, Feature availability, Claude Code documentation. https://code.claude.com/docs/en/feature-availability — consulted 11 October 2026.
3. S3 Official (vendor documentation). Anthropic, Enterprise deployment overview, Claude Code documentation. https://code.claude.com/docs/en/third-party-integrations — consulted 11 October 2026.
4. S4 Official (vendor documentation). Anthropic, Use Claude Code in the cloud, Claude Code documentation. https://code.claude.com/docs/en/claude-code-on-the-web — consulted 11 October 2026.
5. S5 Official (vendor documentation). Anthropic, Automate work with routines, Claude Code documentation. https://code.claude.com/docs/en/routines — consulted 11 October 2026.
6. S6 Official (vendor documentation). Anthropic, Claude Code on mobile, Claude Code documentation. https://code.claude.com/docs/en/mobile — consulted 11 October 2026.
7. S7 Official (vendor documentation). Anthropic, Desktop application, Claude Code documentation. https://code.claude.com/docs/en/desktop — consulted 11 October 2026.
8. S8 Official (vendor documentation). Anthropic, Let Claude coordinate ongoing work with Projects, Claude Code documentation. https://code.claude.com/docs/en/claude-projects — consulted 11 October 2026.
9. S9 Official (vendor documentation). Anthropic, Use Claude Code with Chrome, Claude Code documentation. https://code.claude.com/docs/en/chrome — consulted 11 October 2026.
10. S10 Official (vendor help centre). Anthropic, Getting started with Claude in Chrome, Claude Help Center. https://support.claude.com/en/articles/12012173-getting-started-with-claude-in-chrome — consulted 11 October 2026.
11. S11 Official (vendor documentation). Anthropic, Data usage, Claude Code documentation. https://code.claude.com/docs/en/data-usage — consulted 11 October 2026.
12. S12 Official (vendor documentation). Anthropic, Agent SDK overview, Claude Code documentation. https://code.claude.com/docs/en/agent-sdk/overview — consulted 11 October 2026.
13. S13 Official (vendor documentation). Anthropic, Intro to Claude, Claude Developer Platform documentation. https://platform.claude.com/docs/en/intro — consulted 11 October 2026.
14. S14 Official (vendor documentation). Anthropic, Claude Managed Agents overview, Claude Developer Platform documentation. https://platform.claude.com/docs/en/managed-agents/overview — consulted 11 October 2026.
15. S15 Official (vendor documentation). Anthropic, Models overview, Claude Developer Platform documentation. https://platform.claude.com/docs/en/models/overview — consulted 11 October 2026.
16. S16 Official (vendor documentation). Anthropic, Pricing, Claude Developer Platform documentation. https://platform.claude.com/docs/en/about-claude/pricing — consulted 11 October 2026.
17. S17 Official (vendor price list). Anthropic, Claude plans and pricing. https://claude.com/pricing — consulted 11 October 2026.
18. S18 Official (vendor documentation). Anthropic, Claude in Microsoft Foundry, Claude Developer Platform documentation. https://platform.claude.com/docs/en/build-with-claude/claude-in-microsoft-foundry — consulted 11 October 2026.
19. S19 Official (vendor documentation). Anthropic, Work with Claude Tag. https://claude.com/docs/claude-tag — consulted 11 October 2026.
20. S20 Official (vendor documentation). Anthropic, Claude for M365 overview. https://claude.com/docs/office-agents/overview — consulted 11 October 2026.
21. S21 Official (vendor announcement). Anthropic, "Claude Cowork and chat are now one Claude", Claude blog, 16 September 2026. https://claude.com/blog/cowork-is-now-claude — consulted 11 October 2026.
22. S22 Official (vendor help centre). Anthropic, Claude Cowork and chat are one Claude, Claude Help Center. https://support.claude.com/en/articles/16761823-claude-cowork-and-chat-are-one-claude — consulted 11 October 2026.
23. S23 Official (vendor help centre). Anthropic, Cloud sessions bonus credit promotion, Claude Help Center. https://support.claude.com/en/articles/17152539-cloud-sessions-bonus-credit-promotion — consulted 11 October 2026.
24. S24 Official (vendor help centre). Anthropic, Pair the Claude mobile app with Claude Desktop (Dispatch), Claude Help Center. https://support.claude.com/en/articles/13947068 — consulted 11 October 2026.
25. S25 Official (vendor announcement). Anthropic, "Updates to Consumer Terms and Privacy Policy", Anthropic news. https://www.anthropic.com/news/updates-to-our-consumer-terms — consulted 11 October 2026.
26. S26 Official (vendor privacy centre). Anthropic, Is my data used for model training?, Claude Privacy Center. https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training — consulted 11 October 2026.
27. S27 Primary (user conversations). Two conversations held with Claude on 8 and 10 October 2026 by the author of this knowledge base, exported as Markdown: "Sessions Chat (cloud) vs Claude Code" and "Conversation chat Claude Cloud". Unpublished; quoted in sections 1, 6 and 7 in the author's own words.