New · October 2026
Work acrossfolders
Keep the app, the API and the shared library in one chat. A chat has a main folder, and you can add up to 15 more. The AI searches and edits all of them, reads each one's AGENTS.md, and gets the skills, agents and connectors you set up for each. Adding a folder is the trust step: files there no longer need a permission prompt, so add only what you mean to share. Similar to Claude Code's `--add-dir`, with each folder's own setup coming along.
Related features
Add a Folder From the Pill or the Chips
A chat has one main folder. Open the workspace pill and pick "Add a folder to this chat", or use the chips strip above the composer. Each extra folder becomes a chip with its name on it. The chat lists them under "Also working in".
- Hover a chip to see the full path
- One click on a chip removes the folder from the chat
- Up to 15 extra folders per chat
Adding a Folder Is the Trust Step
Once a folder is on the chat, the AI can read, search and edit files in it without asking, even in safe mode. That is the same trust the main folder already gets. Searches with `plexon_grep`, `plexon_find_files` and `plexon_code_search` cover the main folder and every extra one unless you narrow them.
- No permission prompt for files inside an added folder
- One search covers every folder on the chat
- Each extra folder's AGENTS.md goes into the instructions; its CLAUDE.md, hooks and memory are never read
Each Folder Brings Its Own Code Intelligence and Setup
Code intelligence runs one language server per folder. Ask about a file in an extra folder and the answer comes from that folder's own server, not the main one. Skills, agents and MCP connectors you turned on for a folder are available in any chat that includes it, so two repositories with different setups can share one conversation.
- Go-to-definition and references resolve against the right project
- The chat gets the union of every folder's skills, agents and connectors
- Claude Code's `--add-dir` adds a directory; here the folder's own setup comes with it
Send a Sub-Agent to One Folder, or Into Its Worktree
A sub-agent can be pointed at any folder on the chat with `working_dir`. Give it `isolation: worktree` and it works in its own git worktree of that folder instead, and the result waits in the Parallel Agents review queue with Merge and Discard. The AI can also move the chat's main folder with `plexon_change_directory`.
- Worktree runs need the folder to be the root of a git repository
- Two chats on different folders no longer step on each other: every tool call resolves its folder per chat
- The title bar still shows the app's workspace, not the chat's folders
What working across folders changes
Up to 15 Extra Folders
One main folder and up to 15 more on the same chat, each shown as a chip you can remove with one click.
One Search, Every Folder
Grep, file search and code search cover the main folder plus every extra folder by default.
A Language Server per Folder
A question about a file under an extra folder is answered by that folder's own language server.
Worktrees per Folder
Sub-agents can run in a git worktree of any folder on the chat, and their work lands in the Parallel Agents review queue.
Only AGENTS.md Joins the Instructions
From an extra folder, the AI's instructions take its AGENTS.md and nothing else. Its CLAUDE.md, hooks and memory are never loaded.
Nested Folders Are Dropped
A folder inside the main folder, or inside another extra folder, is dropped: the outer folder already covers it.
Connects to
Nothing here works alone. See the whole map for how the pieces fit together.
Where it goes
This feeds Parallel Agents Workspace.
A sub-agent can run in its own git worktree of any folder on the chat, and its work waits in the Parallel Agents review queue.