New · July 2026
Never breakyour work
Moving to a new assistant usually means building your setup a second time. Plexon reads your Claude Code install where it already sits and brings it across: the agents you wrote, the MCP servers you connected, the hooks, the plans, the memory, and every conversation. Plans land in the project each one was written for and keep the approved or draft status they had. Nothing at the source is moved, changed, or deleted.
The path
Four screens, start to finish
Point it at the folder, pick what to keep, confirm where each project lives, and read the receipt. The whole run is one dialog.
Claude Code project detected
Found 6 custom agents, 4 hooks, 2 mcp servers in this workspace. Import them into Plexon?
Step 1
It offers before you go looking
Open a repository that Claude Code set up and the banner appears at the top of the chat with a count of what it found. Your answer is remembered for that workspace: Skip asks again next time, "Don't ask" is permanent, and Settings can always reopen the wizard anyway.
Source: ~/.claude
Select allClear
Global data
Each plan goes to the project it was written for. 72 of these were never approved, and the button above leaves those out.
Search 255 plans
Per-project memory
Step 2
Pick what to keep, down to the plan
Every component arrives counted, so you know what you are agreeing to before you agree to it. Plans open into their own list: each row shows whether you approved it, which project it came from, its size, and its age. Search the list, or use the quick selects. "All except ephemeral" keeps the plans you approved and drops the drafts you never did.
Map each project to its local directory. Unmapped projects will be skipped.
3/4 mappedMost projects were found automatically. Review the paths below and click Import.
C:\work\acme-storeauto
C:\work\payments-apiauto
C:\work\internal-tools
Not mapped. Click Browse to select folder.
Step 3
Confirm where each project actually lives
Claude Code stores each project under an encoded folder name. Plexon reads the real path out of the sessions themselves and fills most of them in for you. Anything it cannot place waits here for a folder, and a project you leave unmapped is skipped rather than guessed at.
Plans appear in the Plans panel when their project is open.
2026-04-11-a1c9.jsonl: already imported
2026-02-03-77fe.jsonl: no messages to import
legacy-notes.jsonl: malformed JSON at line 40, rest of the file imported
Step 4
A receipt, not a progress bar
The results screen names every destination and every skip. Plans list the project they landed in with a count and a button that opens the folder. Anything the import would not touch says why, file by file, so "3 skipped" is never something you have to go and investigate yourself.
Everything you set up comes with you
The wizard runs in two places. From Settings it reads your global install at ~/.claude and brings over agents, MCP servers, hooks, plans, your CLAUDE.md and user rules, per-project memory, and the full conversation history. Open a repo that Claude Code set up and a second, smaller pass offers that project's own .claude folder without leaving the chat. Ctrl+K opens either one: type claude or migration.
- Global: agents, MCP servers, hooks, plans, global instructions, project memory, sessions
- Per repo: agents, project skills, MCP servers, hooks, slash commands
- Agents arrive as Plexon sub-agents, hooks as project hooks, memory as memory
- Pick component by component, or take the lot with Select all
Plans go to the project they were written for
An import used to drop every plan into one global folder. A plan for your API then sat next to a plan for your store, and all of them were read into the system prompt on every turn. Plexon now works out which project each plan belongs to by reading the session that produced it, then writes a real plan into that project's own store with its title, its original date, and its approved or draft status.
- Attribution measured at 255 of 255 plans on a 2.6 GB history, with no ambiguities
- Two small reads per session file, about 0.6% of the history on disk
- A project folder you have since deleted is still a valid destination
- Plans that cannot be traced go to one folder you choose, not scattered across a dozen
Leave the drafts you never approved behind
Approving a plan leaves a mark in the session history, so the absence of that mark is what makes a plan a throwaway draft. The wizard counts them and gives you one button to drop them. On a real history that is usually a third of the pile.
- "All except ephemeral" keeps the plans you approved and drops the rest
- When the history is gone, sub-agent drafts are still spotted by filename
- A plan the model is writing right now is never counted as a draft
- The scan is cached against file stats, so re-opening the wizard costs about half a second
Sessions land in the project they belong to
Every project keeps its own encrypted session store. The import opens the right store for each project instead of filing a dozen repositories' conversations under whichever folder Plexon happened to start in.
- Sessions keep Claude Code's own ids, so a second run skips instead of duplicating
- A project folder that is not on this machine is reported, not guessed at
- Claude Code's slash-command bookkeeping is filtered out, so imported chats read clean
- Sub-agent transcripts stay out of the main conversation
Slash commands become skills
Plexon has no slash commands. It has skills with trigger phrases, so a .claude/commands/deploy.md arrives as a skill that fires on "deploy" and "/deploy". Nothing is overwritten on the way in: a name that already exists gets a suffix instead.
- Commands convert to a SKILL.md with mandatory triggers and an imported-from stamp
- A skill name that already exists becomes name-claude
- A project MCP server clashing with a global one is suffixed with the workspace name
- Project skills copy across as they are, the format already matches
Everything imported is labelled
One run can add hundreds of sessions and plans. Each one is stamped with where it came from, so you can find the whole import afterwards, read through it, and undo it. Sessions show a Claude Code label in the sidebar. Plans show it on the card and gain an Imported filter that sits next to All, Approved, and Drafts.
- Type "claude" in either search box to pull the whole import up as a group
- The Imported filter appears only once something has actually been imported
- Ctrl+A selects the filtered list, so clearing an import is three clicks
- The label reads as a product name on screen, never as the raw id underneath
Run it twice, nothing doubles
Every part of the import checks for what is already there before it writes. Add three agents next month, run the wizard again, and three agents arrive.
- Agents, skills, MCP servers, and hooks skip on name
- Plans skip on an id derived from the source filename, so the same plan lands once
- Sessions skip on Claude Code's own session id
- Nothing under ~/.claude is moved, edited, or deleted, so Claude Code keeps working
Afterwards
Find them again, or clear them out
A migration can add hundreds of items in one go. Every one of them is marked with where it came from, so the whole import stays reviewable and reversible.
Rework the checkout flow
ApprovedClaude Codeacme-store
Split the pricing service
ApprovedClaude Codepayments-api
Rewrite the docs navigation
ApprovedClaude Codesite
claude
Refactor the auth middlewareClaude Code
acme-store · 41 messages · 3d ago
Pricing service breakdownClaude Code
payments-api · 118 messages · yesterday
Storefront copy passClaude Code
site · 27 messages · 6d ago
Ctrl+A selects the whole filtered list, so removing an import you changed your mind about is a filter, a select, and a delete.
Why the migration matters
Two entry points
Settings migrates the global install at ~/.claude. Opening a repo that Claude Code set up offers that project's own .claude folder on the spot, and remembers your answer for that workspace.
Per-project attribution
255 of 255 plans matched to the project that produced them on a 2.6 GB history, with no ambiguities and no folder guessing. The session's own recorded working directory is the source of truth.
Drafts stay behind
Approval leaves a mark in the session history. One button keeps the plans you approved and drops the ones you never did, instead of importing every scratch file you ever generated.
Labelled for good
Sessions and plans carry the source they came from. Both panels badge it, both search on it, and Plans adds a dedicated Imported filter so a bulk import stays reviewable.
Safe to re-run
Every importer skips what already exists rather than writing a second copy. Sessions and plans carry stable ids, so the list never doubles on a second pass.
Your old setup keeps working
Nothing in ~/.claude is moved, edited, or deleted. The migration reads, Plexon writes its own copies, and Claude Code carries on exactly as it did.