Claude Code Sessions: Self-Hosted Environments and Session Messaging in 2.1.224

Claude Code sessions: start, continue, and resume

The first distinction to make is between starting a conversation, continuing the latest conversation in a directory, and selecting a particular previous conversation. These operations use different commands, so choosing the right one matters when you have several conversations associated with your work.

The official CLI reference documents the basic entry points:

Continue the latest conversation

Use the continue command when your intended conversation is the most recent one in the current directory:

claude --continue

Its short form is:

claude -c

The directory qualification is important. The documented behavior is to load the most recent conversation in the current directory, so “continue” should not be read as a global selector for whichever conversation you last used anywhere.

Resume a chosen conversation

When you want to choose from previous conversations, use:

claude --resume

To select a known session directly, supply its ID:

claude --resume <session-id>

Inside an interactive session, /resume provides a picker. These commands answer different needs: continue selects the latest conversation in the current directory, while resume gives you interactive or explicit selection. The Claude Code CLI reference is a useful companion when choosing between command-line entry points.

Saved conversations, named sessions, and live-session discovery

Session selection and live-session discovery serve different purposes. The resume picker helps you select a previous conversation; the live-session command lists Claude sessions that are currently running.

Keeping that distinction clear is especially useful now that sessions can message one another. Finding a conversation you want to reopen and finding a live session you want to contact are separate tasks.

Name a conversation for later reuse

Named sessions predate Claude Code 2.1.224. Version 2.0.64 introduced support for naming a conversation with /rename and reopening it by name.

Once a conversation has a name, you can select it from the CLI:

claude --resume <name>

You can also use /resume <name> within a session. A name gives you another way to identify a conversation alongside its session ID and the interactive picker.

For example, the choice between continuing and resuming remains the same even when you use names: continue selects the latest conversation in the current directory, while a named resume selects the conversation you specify. Naming does not change the documented selection behavior of --continue.

List sessions that are running

For programmatic discovery of live sessions, use:

claude agents --json

This command, introduced in v2.1.145, lists live Claude sessions as JSON. It is useful when your immediate question is which sessions are running, rather than which previous conversation to reopen.

The version-pinned changelog records both named-session support and the live-session listing command. Neither capability first appeared in 2.1.224; they provide existing session-management tools alongside the new messaging features.

Self-hosted environments in Claude Code 2.1.224

The v2.1.224 release introduces claude self-hosted-runner. It lets organizations use their own machines or containers to execute Claude Code sessions initiated through the web, mobile, and desktop interfaces.

The command named in the release is:

claude self-hosted-runner

Availability is stated for Team and Enterprise plans. Those are the verified plan requirements for this feature as of the publication cutoff.

What the runner changes

The runner gives organizations an execution option on machines or containers they provide. The supported session interfaces listed in the release are web, mobile, and desktop.

That scope is the central point of the Claude Code self-hosted environment announcement. It describes where those sessions can execute and which plans can access the feature. It does not provide a complete deployment recipe in the verified material.

If you are evaluating how this fits your use of Claude Code, the web and desktop guide provides a related place to explore those interfaces. For this release, keep the execution feature, interface coverage, and plan eligibility as separate evaluation points.

What remains unverified

The release does not give a separate runner price, a per-session charge, or a self-hosted session quota. It also does not establish detailed deployment flags or a verified configuration example in the material available for this article.

Those gaps matter when planning a deployment. The documented command establishes the feature’s entry point, but it is not enough to infer a complete setup procedure or a cost model. Any estimate of runner charges or session allowances would go beyond the supplied release information.

Cross-session messaging and inbound approvals

Version 2.1.224 adds cross-session SendMessage, with ListAgents for discovery, on macOS and Linux. The release describes messaging sessions “on any of your machines.”

That wording supports cross-machine messaging in the release description. Windows support is not listed for this feature, so the verified platform coverage here is macOS and Linux.

Discovery and messaging have different roles

ListAgents provides discovery for cross-session messaging, while SendMessage sends a message to another session. These are the named capabilities in the release; the verified material does not provide invocation examples or parameter schemas for them.

The existing claude agents --json command also lists live sessions, but it should not be substituted for an undocumented ListAgents invocation. One is a verified CLI command, and the other is the discovery capability named alongside cross-session messaging.

For developers, this creates three distinct session operations: reopening a saved conversation, discovering live sessions, and sending a message to a session. Use the resume commands for the first task and the documented discovery or messaging capabilities for the others.

Incoming messages can require approval

The release adds two settings, crossSessionInbound and dialogExpiry, alongside messaging. It also specifies an approval distinction:

• Messages addressed to sessions running with bypassed permissions wait for user approval.
• Messages addressed to other sessions automatically deliver.

A message’s approval behavior therefore depends on the receiving session’s permission mode. When coordinating sessions, account for the possibility that an incoming message is waiting for approval before treating it as delivered.

The verified release material does not specify accepted values, defaults, expiry units, or configuration examples for crossSessionInbound and dialogExpiry. Their names alone are insufficient to construct a reliable settings file.

Failed deliveries now surface as errors

The release also fixes a misleading SendMessage result. Previously, it could report “Message sent” even when writing to a teammate’s inbox failed. In 2.1.224, failed deliveries are reported as errors.

This correction matters when interpreting message results. Delivery failure is now surfaced through an error, while the inbound approval rule remains a separate consideration for sessions running with bypassed permissions.

Session isolation, subagent limits, and Remote Control fixes

The headline additions are accompanied by changes to session handling. These fixes affect project isolation, subagent spawning, and Remote Control behavior when resuming work.

Long project paths resolve correctly

Version 2.1.224 fixes project paths longer than 200 characters resolving to another project’s session directory. The affected operations include session listing, renaming, forking, deletion, and /resume.

This is relevant if your project lives under a deeply nested path. The fix covers several session operations, so its scope extends beyond the resume picker alone.

The total subagent spawn cap is removed

The release removes the 200-subagent-per-session spawn cap. Concurrency and depth limits still apply.

Those are different constraints: removing the total spawn cap does not remove the remaining limits on concurrent activity or depth. When assessing a session that uses subagents, keep all three concepts separate. The Claude Code subagents guide is the related reference for that part of a coding workflow.

Resuming respects disabled Remote Control

Resuming a session no longer silently reconnects Remote Control after the user disabled it. The fix covers --resume, SDK hosts, and the VS Code extension.

VS Code also receives a fix for explicitly enabled remoteControlAtStartup being ignored. Together, these changes address both a disabled state being overridden during resume and an explicitly enabled startup setting being ignored in VS Code.

These are fixes to session behavior, not additional commands you need to add to a resume workflow.

Key takeaways

• Claude Code 2.1.224 shipped on August 7, 2026, adding self-hosted runners and cross-session messaging.
• Use claude --continue for the latest conversation in the current directory, and claude --resume to select a previous conversation or specify its ID or name.
• claude agents --json lists live sessions; /resume provides a picker for reopening conversations.
• claude self-hosted-runner supports execution of web, mobile, and desktop sessions on an organization’s machines or containers, with availability stated for Team and Enterprise plans.
• Cross-session messaging is listed for macOS and Linux. Incoming messages to sessions with bypassed permissions wait for user approval, and failed deliveries now report errors.
• The release fixes long-path session isolation and Remote Control resume behavior, and removes the 200-subagent spawn cap while retaining concurrency and depth limits.

FAQ

What is the difference between --continue and --resume?

claude --continue, or claude -c, loads the most recent conversation in the current directory. claude --resume opens interactive selection, while supplying a session ID or name selects that conversation directly.

How do I list saved conversations versus live sessions?

Use claude --resume or the in-session /resume picker to select a previous conversation. Use claude agents --json to list live Claude sessions as JSON; that command was introduced in v2.1.145.

Which plans support self-hosted Claude Code environments?

The v2.1.224 release states that self-hosted runners are available on Team and Enterprise plans. The runner lets organizations execute web, mobile, and desktop sessions on their own machines or containers, but the release gives no separate runner price or self-hosted session quota.

Can Claude Code sessions message each other across machines?

Version 2.1.224 adds cross-session SendMessage and ListAgents discovery on macOS and Linux, describing messaging sessions “on any of your machines.” Windows support is not listed, and messages addressed to sessions running with bypassed permissions wait for user approval.