Skip to main content

Guides

Separate Work and Personal Claude Accounts

Separate work and personal Claude accounts with named Power Claude groups

Need to separate work and personal Claude accounts on one machine without mixing transcripts, memory, or project history? This step-by-step guide shows how to name Power Claude groups, assign any seats, isolate context, and switch the active group (dashboard + CLI).

TL;DR

StepAction
1Create groups named Work and Personal (isolation on)
2Drag each login into the correct group (any linked account can go in either)
3Click Use this group (or pc scope use) whenever you switch roles
4Confirm separate trees under ~/.power-claude/claude-shared/scopes/
5Optional: add more named groups for clients, projects, or sandbox
Full reference: [Account groups deep dive](/blog/guides/power-claude-account-groups-isolated-context) · [Setup docs](/docs/getting-started/account-groups).

Why separating work and personal Claude accounts needs more than switch

Switching the active Claude login changes who pays / which seat answers. It does not automatically give you:

  • A named set of seats that belong to “work” vs “home”
  • A hard boundary so work memory never lands in personal sessions
  • A durable “I am in personal mode now” selection you can re-apply tomorrow

Without groups, people improvise: sticky notes, shell aliases, or hoping the last account was the right one. That fails the first time a personal experiment inherits enterprise conventions — or a work agent sees side-project clutter.

Account groups add what manual switching and flat rotators lack: a named set of seats for the engagement, membership (only those accounts), pool that set while it is active, and optional isolation of projects/memory. Work is not “every login on the laptop”; it is the Work group only.

Before you start

  • Power Claude installed; at least one Claude account linked under Your Profiles.
  • Prefer official-cli / normal Claude Code logins per profile (Architecture B).
  • Decide names. Work / Personal is enough for most people. You can also use Day-job / Side-projects — names are free-form.
  • If you only have one login today, still create both groups so the second seat has a home when you add it.

Step 1 — Name the groups

Dashboard

  1. Power Claude → Your ProfilesAccount groups.
  2. + New group → type Work → create (isolation defaults on).
  3. + New group → type Personal.

CLI

pc scope create Work
pc scope create Personal
pc scope list

You can name groups anything later for clients (Client-Acme) or projects (Project-Alpha). Stable short names work best with shell aliases and direnv.

Step 2 — Put accounts in the right groups

Dashboard

Drag each account chip from Ungrouped onto Work or Personal. Any linked account can join any group — there is no special account type.

CLI

pc scope add <work-scope-id> you+work@company.com
pc scope add <personal-scope-id> you@personal.com

Rule: one account → one group. That keeps isolation unambiguous. For multi-label without dual isolation (e.g. “favorite” + “backup”), use tags.

Move mistakes

pc scope move you@personal.com <personal-id>
pc scope move you@personal.com none   # back to Ungrouped

Step 3 — Activate the persona you are right now

MomentAction
Starting workUse this group on Work / pc scope use <work-id>
Side project after hoursUse Personal
Switching mid-dayUse the other group before opening that repo
Done for the dayClear active group / pc scope use none
Activating applies the group’s projects layout (isolated tree) without touching credentials. OAuth stays private per account.

Step 4 — Verify separation

pc scope list
ls ~/.power-claude/claude-shared/scopes/

Practical check:

  1. Use Work → leave a short throwaway note or distinctive memory in a work session.
  2. Use Personal → open Claude Code.
  3. Confirm you are not continuing the Work memory tree.

If both personas still share history, re-check isolation is isolated (not global) and that Use this group actually ran after membership changes.

Step 5 — Optional: more groups (clients, projects, sandbox)

The same create → name → assign → use loop scales:

Group nameTypical membersWhen to activate
Client-AcmeSeats used only for AcmeBefore Acme monorepo
Client-BetaSeats for BetaBefore Beta work
Project-Q3Seats billed / tracked to Q3 launchDuring that engagement
SandboxThrowaway experimental seatsRisky agent runs
You can **group any accounts** into these sets and **name them** for how you actually work — not only Work/Personal.

Decision: groups vs tags vs config dirs

ToolJob
CLAUDE_CONFIG_DIR / official profilesSeparate logins
Account groupsSeparate personas + local memory + active set
TagsSoft labels only — no isolation
Use profiles so credentials do not collide. Use groups so work and personal brains do not collide. Use tags when you only need a label.

Deep dive: Account groups: full guide · Official multi-account vs relay.

Common mistakes

MistakeFix
Only tags, no groupsTags do not isolate memory — create Work/Personal groups
Isolation set to global for WorkUse isolated unless you intentionally want shared memory
Personal email still in Workpc scope move you@personal.com <personal-id>
Created groups but never activatedUse this group or pc scope use before deep work
Expecting double quotas from groupsGroups do not multiply Anthropic limits
One giant “Everything” groupSplit on persona or billing boundaries, not seat count
Renaming by editing JSON by handPrefer UI/CLI; backup account-scopes.json first

Daily habit (keeps separation real)

  1. Morning → Use Work.
  2. Work repos only under that group.
  3. Evening / learning → Use Personal.
  4. Client day → Use that client group before opening the repo.
  5. Optional: wire pc scope use <id> into direnv per checkout.

The habit matters more than the perfect initial naming. You can always create another group and move accounts.

Frequently asked questions

Is this the same as CLAUDE_CONFIG_DIR?

Related but different. Config dirs separate logins; groups separate persona + local projects/memory and the active set. Use both for a clean setup.

Can I rename a group later?

Prefer a stable name from day one. If rename is not in the UI, create the correctly named group, move members, delete the empty old group. Keep account-scopes.json backed up.

What about only one account?

You can still create Work and Personal groups for future seats, or skip groups until you link a second login. Isolation pays off as soon as a second persona exists.

How do I group more than two personas?

Same flow: name a group per persona (or client/project), drag any accounts in, activate before work in that role.

Can I put any account in any group?

Yes. Membership is free-form. Drag (or pc scope add) any linked login into any named group.

Do groups share OAuth or passwords?

No. Credentials stay private per account. Groups never copy secrets.

Will this delete my old Claude sessions?

No. Isolation re-homes context roots; it does not wipe history. Layout apply fails closed rather than clobbering a non-empty tree.

Does separating work and personal bypass rate limits?

No. Limits remain per Anthropic account/plan. Groups organize seats and local memory only.

How do I clear the active group?

Dashboard Clear active group or pc scope use none.

CLI and dashboard disagree — which wins?

They should not disagree: both write ~/.power-claude/state/account-scopes.json. Reload the dashboard or re-run pc scope list if a view looks stale.

Can I use this for project billing boundaries?

Yes as local organization: name a group per project or client, put the seats you use for that work inside it, activate while on that engagement. That aligns context with the work; it does not replace Anthropic billing.

Where do I read the full model?

Power Claude account groups deep dive and docs: account groups.

Setup checklist

  1. Link work and personal Claude accounts.
  2. Create isolated groups named Work and Personal (or your preferred names).
  3. Drag each login into the correct group.
  4. Use Work before work sessions; Use Personal before personal sessions.
  5. Verify with pc scope list and a quick memory-tree check.
  6. Add client/project groups when you need more boundaries.
  7. Optional: pc scope use in direnv per repo.

Related

Independent third-party software. Not affiliated with Anthropic.