Compare·Provider
Continuum
Multi-agent workbench
VS
Claude Code
Official Anthropic coding CLI

Claude Code in Continuum

Same Claude Code quality. Continuum adds the multi-device workbench, gauges, and multi-lab peers.

Claude Code is Anthropic’s official agentic coding CLI - plan mode, tools, and Max subscription rails. Continuum is not a fork and not a substitute model. It drives real Claude Code under your login, then adds every surface, every other provider, multi-account isolation, and a local cost ledger around it. Continuum’s spawn path uses the real `claude` binary and PTY, not a reimplemented Anthropic agent. Plan mode, tools, and Max rails remain Anthropic’s; Continuum owns the multi-device projection and multi-account isolation around them.

Updated 2026-08-03 · Mac stable · Win/Linux desktop beta

Choose Continuum when

You want Claude Code plus Codex/Cursor/others, mobile approvals, multi-account Max stacking, worktree Code panes, and analytics - without living only in a terminal pane.

Choose Claude Code when

You want the purest Claude Code CLI experience and nothing else - one terminal, one vendor, no extra workbench.

Snapshot provider · Continuum hosts it
Dimension Continuum Claude Code
Engine identity Hosts official Claude Code (PTY) Is the official Claude agent CLI
Auth path Your Anthropic / Max login · no proxy markup Claude plan / setup-token / CLI login
Multi-account CLAUDE_CONFIG_DIR / CODEX_HOME isolation One login per machine by default
Surfaces Continuum adds Mac · web · iOS · Watch · Code panes Terminal / Claude’s own surfaces
Gauges Menu-bar 5h + weekly gauges CLI / account usage views
Analytics Local JSONL → $ by repo/provider/day Anthropic account usage, not repo ledger
Mac stable · web · iPhone · Watch · Win/Linux desktop beta · free app
01 · Engine in the terminal vs cockpit across devices
In Continuum

Claude day with cockpit ops

08:50

Mac menu-bar gauge shows the 5h window. You open Continuum Code and spawn Claude in a worktree for the morning ticket.

13:10

Plan mode finishes. Approve from iPhone. Diff and PR panes stay on the same session thread.

19:20

Primary Max is hot. Pin the secondary Claude account for the night batch; analytics still roll into one ledger. Spawn grid can open plain `claude` TUI tiles when you want interactive CLI beside managed worktree sessions.

In Claude Code

Terminal Claude only

08:50

Open a terminal. Run `claude` in the repo. Plan mode and tools work exactly as Anthropic ships them.

13:10

Stay at the keyboard for approvals and tool confirms. No Continuum phone client on this path.

19:20

One machine login. When the window is hot, you wait or rearrange accounts yourself outside a multi-instance workbench. Pure `claude` remains perfect for quick one-off terminal work with zero Continuum chrome.

02 · Monday path

Monday on Claude Max with Continuum as cockpit

Install, connect, first session - steps you can run the same day.

01

Install Continuum Mac; confirm `claude` is on PATH and your Max login works.

02

Open Code tab → new session → Claude agent → worktree mode for the morning ticket.

03

Watch the menu-bar 5h gauge before a heavy run; Usage tab shows weekly heat.

04

When plan mode finishes away from the desk, approve from iPhone on the paired host.

05

Mid-day: open a Codex peer session on the same repo if you want a second opinion branch - Claude stays primary.

06

If the primary window is hot, pin a secondary Claude account (CLAUDE_CONFIG_DIR isolation) for the evening batch.

07

After first Claude session, confirm menu-bar gauge ticks - if it does not, fix auth before stacking a second account.

How Continuum actually runs Claude Code

How Continuum actually runs Claude Code

  • Continuum launches official `claude` over a real PTY with subscription-safe env (no HOME swap).
  • Plan mode, tools, and transcript stream into Code panes (chat, plan, diff, PR, terminal) from the live session.
  • Secondary accounts use CLAUDE_CONFIG_DIR isolation + per-instance Keychain - never clobber primary auth.
  • Live gauges poll Anthropic rate-limit rails; analytics price ~/.claude projects JSONL with LiteLLM rates.
  • iPhone/Watch pair to the Mac daemon over Tailscale / MagicDNS for approve, interrupt, and glance.
  • Analytics walk ~/.claude projects JSONL (and secondary instance dirs) and price them with LiteLLM rates.
03 · Claude Code host matrix

Same engine under your login. Continuum cells name the cockpit - panes, gauges, isolation - not a re-score of model quality.

Capability Continuum Claude Code
Official Claude Code agent Drives real CLI / PTY session Is the CLI
Plan mode + approve Workbench + mobile approve Native plan mode
Multi-provider (Codex, Grok…) First-class peers Claude only
Stack 2+ Claude accounts CLAUDE_CONFIG_DIR / CODEX_HOME isolation One login per machine
iPhone / Watch steering Native No Continuum-style apps
Worktree sessions + PR pane Chat · plan · diff · PR · terminal · artifacts You arrange git yourself
Live 5h / weekly gauges Menu bar + Usage tab CLI / account usage
Local $ by repo Local JSONL → $ by repo/provider/day Not a repo ledger
Spawn grid multi-tile 4/6/8 agent CLI tiles Single terminal focus
Workbench app fee Free app · optional Plus/Max/Ultra hosted Included with Claude plan
04 · What Continuum layers on without forking
01 · Layering

Engine versus cockpit.

Claude Code is the agent. Continuum is the workbench that runs Claude Code (and peers). Comparing them as mutually exclusive products misses the point - Continuum is how you operationalize Claude Code across devices, accounts, and other labs. Continuum’s Code panes (chat, plan, diff, PR, terminal, artifacts) attach to the same Claude session the terminal would have run alone.

02 · Multi-account

Two Max plans, one instrument.

Continuum isolates secondary Claude accounts so you can pin sessions and flip when a window is hot. The raw CLI is one login per machine. That alone is why power users adopt Continuum without leaving Claude. Secondary Max accounts are isolated with CLAUDE_CONFIG_DIR + per-instance secrets - Continuum refuses wrong-account billing fallbacks when pins are invalid.

03 · Fleet

Claude is not the only good agent.

Some tasks prefer Codex, Cursor, or Grok. Continuum keeps Claude Code as a peer, not a monopoly - open a second session on another provider, compare the diffs, keep the better branch. Claude Code alone stays Claude-only by design.

04 · Away-from-desk

The plan lands; you are not at the Mac.

Claude Code in a terminal expects you to be there. Continuum’s mobile clients exist so approve / interrupt / steer can happen from the phone without SSHing into a pane.

05 · Coexistence
Stack recipe

Same engine. Bigger cockpit.

This is not a fork fight. Continuum spawns and steers real Claude Code (PTY) under your Claude login. Usage hits your Max plan. Continuum never proxies Anthropic tokens for those sessions. Stack recipe: keep `claude` naked in a terminal for quick one-offs; run Continuum for multi-device, multi-account, multi-provider days.

06 · Host Claude Code or stay CLI-only
Continuum

Choose Continuum to operationalize Claude Code

  • Host real Claude Code with Mac, web, iPhone, and Watch clients. Continuum Code panes stay first-class: chat, plan, diff, PR, terminal, artifacts.
  • Stack multiple Claude accounts without clobbering Keychain auth.
  • Sit Claude beside Codex, Cursor, and other agents in one sidebar.
  • Watch live 5h/weekly gauges and repo-level spend.
  • Keep the free Continuum app on top of your Anthropic plan.
Claude Code

Stay on raw Claude Code when the terminal is enough

  • Want only Anthropic’s official CLI with zero extra UI. Continuum will not pretend to win that job with marketing wording alone.
  • Prefer a single terminal pane and one Claude login.
  • Do not need multi-provider compare or mobile Continuum apps.
  • Already orchestrate git/worktrees outside any workbench product.
  • Value minimal tooling surface over fleet ops features.
07 · What you pay
Continuum

App + your labs

Free app with BYOK / your subscriptions. Optional Continuum-hosted inference: Plus $25/mo, Max 100, Max 200, Ultra. Claude sessions bill your Anthropic plan.

Claude Code

Their bill

Claude Pro / Max billed by Anthropic (plan tiers change - verify claude.ai / Anthropic pricing as of 2026). CLI is included with the plan that unlocks Claude Code.

How to compare

Total cost of work

As of 2026 - verify on continuumcode.ai/pricing and the vendor site. Continuum does not mark up BYOK or subscription sessions it drives under your login. Continuum does not mark up Claude subscription sessions. It drives the official CLI under your login.

Continuum ladder: Free app · Plus $25/mo · Max 100 · Max 200 · Ultra - full pricing. Also analytics, multi-account, devices.

08 · Questions

Claude Code + Continuum.

Deep dives: docs, providers, sessions.

It is a Claude Code host and workbench - plus every other provider. If you only want Anthropic’s CLI with no extra UI, use Claude Code alone. If you want mobile, multi-account, multi-provider, and analytics, add Continuum.

No. Continuum drives the official CLI under your login. Tokens for those sessions do not route through Continuum as a billed proxy.

Yes for Claude sessions. Continuum does not replace Anthropic billing. The Continuum app is free; optional Continuum-hosted models are separate.

Yes. Pair the iPhone app and you get the same thread - chat, plan, diff, PR - against the session running on your Mac or enrolled host.

No. Continuum does not reimplement Anthropic’s agent. Claude Code remains the engine; Continuum is the multi-device cockpit around it.

Continuum isolates secondary Claude instances with CLAUDE_CONFIG_DIR (and per-instance Keychain partitions) - not a full HOME override that would break git/ssh. Primaries stay on the default machine login path.

Code-mode Continuum sessions skip approvals by design for agent coding velocity; Plan mode does not bypass. Continuum still productizes plan approve on mobile when you are in plan. Check current Continuum permission docs for the two-state model.

Begin

Run Claude Code
from Continuum.

Free app. Your subscriptions. Optional hosted inference. Mac stable - Windows and Linux desktop are beta.

vendor-neutral · local-first · multi-device