VS Code 1.121 Ships Remote Agents Over SSH, Built-In Mermaid Diagrams, and a New Open Agent Protocol
VS Code 1.121, released May 20, adds remote agent sessions over SSH and dev tunnels, native Mermaid diagram rendering, HTML file preview, and a new Agent Host Protocol for multi-client agent coordination.
Overview
Microsoft shipped Visual Studio Code 1.121 on May 20, 2026, with a headline feature set aimed squarely at developers running long-lived AI agent workloads. According to the official release notes, “this release adds built-in Mermaid and HTML previews, streamlines terminal tool behavior for agents, and lets you run agent sessions on remote machines.”
Remote Agents and the Agent Host Protocol
The most significant addition is experimental support for remote agent sessions. The VS Code Agents window documentation describes the capability directly: “You can connect to a remote machine to start a session there or track the progress of an existing session running on that machine.”
Connectivity works via two methods: SSH and dev tunnels. For SSH connections, “the remote machine must be accessible over SSH. No extra agent installation is needed on the remote machine” — VS Code downloads and installs the CLI on the target automatically. For dev tunnels, the documentation notes that tunnels must require authentication via a GitHub or Microsoft account.
The engineering story behind remote agents is the new Agent Host Protocol (AHP). According to the release notes, AHP is an open protocol that enables “coordination of agent sessions across multiple clients simultaneously.” The protocol manages authoritative state through pure reducers and synchronizes it across every connected client. Because AHP is an open specification, third parties can build clients that connect to the VS Code CLI’s agent host, or build AHP-compatible agent hosts that VS Code can connect to.
A key operational detail: “the remote agent host is a long-lived process. Running sessions continue to run on the remote even if your client disconnects.” This means lengthy refactoring or code generation tasks can run on a powerful remote server, with the developer monitoring progress from any device — including via the Agents window’s web client at insiders.vscode.dev/agents.
The documentation further clarifies that “the Agents window and the main VS Code window share the same underlying agent sessions (Copilot CLI, Copilot cloud, and Claude agent),” making it possible to “run and track multiple sessions in parallel across your projects without opening each workspace in a separate window.”
Built-In Mermaid Diagrams and HTML Preview
Two built-in preview features that previously required third-party extensions are now included by default.
The new Mermaid Markdown Features extension ships inside VS Code and renders Mermaid diagrams in the Markdown preview, in notebook Markdown cells, and in chats, using fenced code blocks. Per the release notes, “rendered Mermaid diagrams also support panning and zooming, which makes larger diagrams easier to inspect without leaving the preview.”
For HTML, developers can now preview local HTML files by right-clicking them in the File Explorer or editor tab and selecting “Open in Integrated Browser,” eliminating the need to install a separate extension.
Model Configurability
VS Code 1.121 adds two new settings for routing lightweight background tasks to different models. The chat.utilityModel setting overrides the model for general utility flows — titles, summaries, commit messages, rename suggestions, prompt categorization, and intent detection. A companion setting, chat.utilitySmallModel, targets fast and lightweight utility flows; the release notes recommend “a fast and inexpensive model” for that setting.
These settings give teams running Copilot Business or Enterprise finer-grained control over cost and latency for background AI tasks without changing the model used for primary coding assistance.
Terminal Optimizations for Agent Workloads
Several terminal improvements in this release address the token and latency overhead of agentic workflows. A new VSCODE_AGENT environment variable is set for agent-issued terminal commands, which can be read by CLI tools to enable machine-readable output and suppress animations or progress bars that waste context window space.
The chat.tools.compressOutput.enabled setting now covers a broader range of tool categories: test runners (pytest, jest, cargo test), build tools, linters, Docker commands, and package managers. Background terminals automatically dispose after a command completes while preserving their output in the chat UI.
Security also receives attention: password, passphrase, PIN, and verification-code prompts in the terminal are intercepted, preventing agents from accidentally capturing or replaying credentials during automated operations.
What We Don’t Know
The Agent Host Protocol is under active development and has not yet reached a stable specification. The protocol’s exact versioning scheme, long-term API surface, and compatibility guarantees across VS Code versions remain unspecified in the current release notes. Remote agent support and the Agents window itself remain in preview, meaning breaking changes may occur in future weekly releases.