What Is Claude Code? Features, Plans, Surfaces and History: Plans, Features and Dates in One Place

Claude Code is the agentic coding tool from Anthropic that reads a codebase, edits files, runs commands and integrates with development tools, from the terminal, IDE extensions, a desktop app and the browser. Anthropic released it as a limited research preview on 24 February 2025 and made it generally available on 22 May 2025. This page sets out which plans include it and at what price, the models and aliases it uses, its features and permission modes, supported platforms, admin controls, data handling and a dated release history, all from Anthropic's own documentation, pricing page and announcements. It is written for developers, engineering managers and platform owners deciding how to roll Claude Code out. Last checked: 2026-11-15.

This is an independent guide from NHI Mgmt Group, based on Anthropic's public documentation as of November 2026. It is not sponsored or reviewed by Anthropic. Product features change, so confirm settings against Anthropic's current documentation.

Key takeaways

• Included in paid Claude plans: Pro ($20 monthly or $17 per month billed annually), Max ($100), Ultra ($200), Team seats and Enterprise include Claude Code; the Free plan does not. It can also run on Anthropic Console (API) billing and through third-party cloud providers.
• Five main surfaces: Anthropic states that the terminal CLI, VS Code, JetBrains, the desktop app and the web share the same engine, so CLAUDE.md files, settings and MCP servers carry across surfaces.
• Model aliases, not fixed names: aliases such as opus, sonnet, haiku, fable and opusplan resolve to different models by provider; on the Anthropic API the default is Opus 5.5.
• Six permission modes: Manual, Edit automatically, Plan, Auto, dontAsk and Bypass permissions decide what runs without asking, and managed settings can disable Auto and Bypass for an organisation.
• Retention depends on account type: commercial accounts have a standard 30-day retention period and no training use under commercial terms; consumer accounts choose whether data is used for training (5-year retention if on, 30 days if off).

What is Claude Code?

Anthropic's documentation describes Claude Code as an agentic coding tool that reads your codebase, edits files, runs commands and integrates with your development tools. Anthropic states that it can plan a change, write code across multiple files, run tests, stage changes, write commit messages, create branches and open pull requests. The CLI can take piped input and run non-interactively with claude -p, which is how it is used in scripts and CI pipelines.

Claude Code runs locally by default and sends prompts and model outputs to the model provider over TLS 1.2 or later. Cloud sessions run the same engine in Anthropic-managed virtual machines, or in a self-hosted environment an organisation controls. The Claude Agent SDK exposes the same tools for teams building their own agents.

Key facts at a glance

As of 15 November 2026, from Anthropic documentation, pricing and announcements.

Plans and pricing

Prices are in USD as listed on Anthropic's pricing page on 15 November 2026, excluding tax. Anthropic states that prices may change.

• API and cloud billing: with an Anthropic Console account, or through Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry, usage is billed per token by that provider rather than through a subscription.
• Shared limits: Anthropic states that cloud sessions share rate limits with all other Claude Code usage on the account.
• Desktop app: requires a paid subscription.
• Long context: on Pro, Opus 4.6 with a 1M token window requires usage credits; it is included on Max, Team and Enterprise.

Models available

Claude Code selects models through aliases. The table shows what each family alias resolves to by provider, from the model configuration page on 15 November 2026.

• Other aliases: fable resolves to Fable 5.1 unless overridden (Fable 5 in Claude apps gateway sessions); best uses the Fable model where available, otherwise Opus; opusplan uses Opus in plan mode and Sonnet for execution; sonnet[1m] and opus[1m] request a 1M token window. Anthropic states Fable models are not the default on any plan or provider.
• Context window: Fable 5.1, Fable 5, Sonnet 5 and later, Haiku 5.5 and Opus 4.7 and later run with a 1M token window by default. CLAUDE_CODE_DISABLE_1M_CONTEXT=1 keeps sessions at 200K.
• Effort levels: low, medium, high, xhigh and max on current models (Opus 4.6 and Sonnet 4.6 have no xhigh). Opus 5.5, Sonnet 5.5 and Haiku 5.5 default to medium; set with /effort, --effort or CLAUDE_CODE_EFFORT_LEVEL.
• Organisation allowlists: an admin can set an organisation default model and an allowed models list, and fast mode is refused when it would switch to a model outside that list.

For model details, see the Anthropic Models page.

Features and modes

• Project instructions and memory: a CLAUDE.md file is read at the start of every session; Claude Code can read an existing AGENTS.md instead, and builds auto memory across sessions.
• Skills and hooks: skills package repeatable workflows as slash commands; hooks run shell commands before or after actions such as file edits.
• Subagents and background agents: a lead agent can spawn subagents for parts of a task, and background agents run several full sessions in parallel.
• MCP: Model Context Protocol servers connect Claude Code to external tools and data. See the MCP Security Guide.
• Scheduled tasks: cloud scheduled tasks (formerly routines) run in the cloud on a schedule, on API calls or on GitHub events; desktop scheduled tasks run on the local machine; /loop repeats a prompt in a CLI session.
• Moving between surfaces: Remote Control continues a local session from a phone or browser, Dispatch sends a task from a phone to the desktop app, claude --teleport pulls a cloud session into the terminal and /desktop hands a terminal session to the desktop app.
• CI and review: GitHub Actions and GitLab CI/CD integrations automate review and triage, and a GitHub Code Review option reviews every pull request.

Permission modes decide what runs without a prompt:

Deny rules apply in every mode, including Bypass. Writes to protected paths such as .git, .claude and .bashrc are never auto-approved outside Bypass. Cloud sessions offer Accept edits, Plan and Auto.

Supported IDEs and platforms

On native Windows, Git for Windows is recommended so Claude Code can use its Bash tool; without it Claude Code uses PowerShell. The terminal CLI, VS Code and JetBrains also support third-party model providers.

Choosing a surface is mostly about where the work runs and who watches it:

• Local and interactive: the terminal CLI or an IDE extension, when a developer wants to review each diff and approve commands on their own machine, with local files and tools.
• Visual review and parallel sessions: the desktop app, which reviews diffs visually, runs several sessions side by side and schedules recurring local tasks.
• Long-running or remote work: the web and mobile surfaces, which run cloud sessions on repositories that are not cloned locally and keep running when the laptop is closed.
• Unattended automation: claude -p in GitHub Actions or GitLab CI/CD, or cloud scheduled tasks, where no person approves each step, so permission rules and the CI identity's access carry the control.

Enterprise and admin controls

• Managed settings: administrators can set permissions.disableBypassPermissionsMode and permissions.disableAutoMode to "disable", and set permissions.defaultMode as the organisation's starting mode for terminal sessions.
• Auto mode defaults: Auto is available by default on Team and Enterprise plans and is the built-in starting mode in interactive sessions from v2.1.283; organisations with the HIPAA configuration start in Manual.
• Model governance: organisation default model and allowed models list (see Models available).
• Zero data retention: available to qualified accounts for Claude Code on Claude for Enterprise, enabled per organisation by the account team; it is not part of the standard Enterprise plan.
• Self-hosted environments: organisations can route cloud sessions to infrastructure they control.
• Monitoring: usage metrics can be sent to an organisation's own OpenTelemetry collector.

For organisation-level controls across Claude products, see the Claude for Enterprise Security Guide.

Data handling and privacy

• Local transcripts: stored in plaintext under ~/.claude/projects/ for 30 days by default; change with cleanupPeriodDays.
• Feedback: transcripts sent with /feedback, /bug or /share are retained for 5 years; disable with DISABLE_FEEDBACK_COMMAND=1. Transcripts shared after a session survey are retained for up to 6 months.
• Telemetry: metrics exclude code, prompts and file paths; turn off with DISABLE_TELEMETRY=1, error reports with DISABLE_ERROR_REPORTING=1, or all non-essential traffic with CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC. These are off by default on Bedrock, Agent Platform, Foundry and Claude Platform on AWS.
• Cloud sessions: the repository is cloned into an isolated VM; GitHub credentials are stored encrypted on Anthropic's servers and attached by a proxy, never entering the VM; outbound traffic passes a security proxy for audit logging.
• WebFetch check: the hostname (not the full URL) is checked against an Anthropic blocklist on every provider; skipWebFetchPreflight disables it.

Release history

Limits and known constraints

• Plan coverage: not included in the Free plan; the desktop app and --teleport and /desktop handoffs need a claude.ai subscription.
• Shared rate limits: cloud sessions draw on the same limits as local use.
• Provider differences: aliases resolve to older models on some third-party providers, so the same alias can mean different models across teams.
• Mode restrictions: settings files other than user and managed settings ignore auto and bypassPermissions as a default mode; cloud sessions do not offer Bypass.
• Update channels: Homebrew and WinGet installs do not auto-update, so versions drift unless upgraded.

Security guidance

Authentication, admin roles, API tokens, sandboxing and logging for Claude Code are covered in the Claude Code Security Guide. For API key handling, see the Anthropic API Keys Security Guide, and for agent risks in editors and terminals, see the AI IDE and Coding Agent Security Guide.

Practitioner checklist

• Pick the billing route first: subscription seats, Console API billing or a cloud provider, since this changes models, telemetry defaults and retention terms.
• Use commercial plans for company code so commercial training and retention terms apply.
• Deploy managed settings that disable Bypass permissions and set the organisation's default mode before rollout.
• Decide whether Auto mode suits your teams, and disable it with managed settings where it does not.
• Set an organisation default model and allowed models list, and record which model each alias resolves to on your provider.
• Shorten cleanupPeriodDays where local plaintext transcripts are a concern, and set telemetry variables to match policy.
• Review MCP servers, hooks and skills committed to repositories as you would any code.
• Standardise on the native installer or manage upgrades for Homebrew and WinGet installs.

Standards and references

• Anthropic: Claude Code overview
• Anthropic: Claude Code data usage
• Anthropic: Claude Code model configuration
• Anthropic: Claude Code permission modes
• Anthropic: Claude pricing
• Anthropic: Claude 3.7 Sonnet and Claude Code
• Anthropic: Introducing Claude 4
• Anthropic: Claude Code on the web

Related NHI Mgmt Group resources: Claude Code Security Guide, Claude for Enterprise Security Guide, Anthropic API Keys Security Guide, MCP Security Guide.