Cortex Code Desktop v1.16.0 release notes¶
v1.16.0 adds SSH remote workspaces in Agent Mode, a new Personalization and Memory experience, built-in web search, and a refreshed Agent Manager, along with reliability and stability improvements.
What’s new¶
-
SSH remote workspaces in Agent Mode: Connect to remote SSH hosts and work with them as first-class projects. File operations, search, git, and terminals all run on the remote host.
-
Personalization page: A dedicated Agent Settings tab with toggles for web search, web fetch, and notebook tools, plus custom instructions backed by
~/.snowflake/cortex/AGENTS.md. -
Memory: Enable, disable, reset, and browse memories, with global and project-level
MEMORY.mdapplied automatically at the start of each conversation. -
Built-in web search: Web search powered by Brave, available directly to the agent.
-
Refreshed Agent Manager: A collapsible sidebar with a
Cmd+Btoggle, Apps promoted to a first-class page, and a new Settings footer. -
Changes panel: A unified diff and changed-files view with a scope selector.
-
Workspace-scoped SQL Playground sheets: Keep multiple
.sqlworksheets per workspace, with rename and drag-and-drop. -
Context window breakdown: Hover the composer to see how your context window tokens are being used.
-
MCP connector improvements: Connector status now reflects real availability, with per-connector Stop and Restart actions.
-
Plan Mode improvements: Newly created plans activate automatically, so updates flow straight to the plan card.
-
Multiple skill commands per message: Chain several
/skillcommands in one message and they all run. -
Send queued messages with double-Enter: Press Enter once to queue a message mid-run, and again to send it right away.
Reliability and fixes¶
- Bash tool hangs: No more hangs on heredocs and multi-line commands, including
gh pr create. - Single diff pill: File edits now show a single persistent diff pill instead of duplicate rows.
- No rogue auto-links: Plain URLs in chat are no longer auto-linkified, so code blocks and SQL identifiers render correctly.
This release also includes additional reliability and stability improvements, including more robust agent and subagent requests, improved dbt command execution and cancellation, and persistent model selection across restarts.