Compare·Self-hosted personal AI assistant
Continuum
Multi-agent workbench
VS
OpenClaw logo
OpenClaw
Self-hosted personal AI assistant

Continuum vs OpenClaw

One is reachable from every chat app you already use. The other is built around the branch, the diff, and the pull request.

OpenClaw is a self-hosted personal AI assistant that started as Peter Steinberger’s Clawdbot in November 2025, was renamed twice, and is now MIT licensed under the nonprofit OpenClaw Foundation with over 380,000 GitHub stars as of August 2026. Architecturally it is one Node.js gateway process that owns sessions, routing, and channel connections, bridging Discord, Slack, Telegram, WhatsApp, Signal, iMessage, Matrix, Teams, and more to an agent that can run shell commands, write code, and touch your GitHub repos, with native macOS, iOS, and Android companion nodes for Canvas, camera, and voice. Continuum is not a general assistant. It is a coding workbench: official Claude Code and Codex CLIs under your own subscriptions, a git worktree per task, per-hunk diff staging, a PR pane, live 5h and weekly plan gauges, multi-account isolation, and a local dollar ledger by repo. Both run on your machine. Only one is shaped like a repo. Context: agentic coding.

Download for Mac All comparisons
Workflow Monday path Matrix Decide Money FAQ

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

Choose Continuum when

Your agent’s output is code that has to be reviewed and merged: you want worktree isolation, a diff you stage by hunk, a PR pane, quota gauges, and spend by repo.

Choose OpenClaw when

You want one always-on assistant you can message from WhatsApp, Telegram, or Slack that handles life admin, automations, and the occasional coding errand on your own hardware.

Snapshot adjacent job
Dimension Continuum OpenClaw
Job-to-be-done Supervised multi-provider coding Always-reachable personal assistant
Control surface Native Mac, web, iPhone, Watch clients Chat apps: WhatsApp, Slack, Telegram, iMessage
Architecture Local daemon + paired clients over Tailscale One Node.js gateway bridging channels
Coding depth Worktrees · diff staging · PR pane · official CLIs Runs shell, writes code, touches repos
Review gate Plan mode, checkpoints, per-hunk staging You trust the thread
Quota + cost 5h/weekly rails · $ by repo/provider/day Your own model metering
Governance Free app, commercial product MIT, nonprofit OpenClaw Foundation
Mac stable · web · iPhone · Watch · Win/Linux desktop beta · free app
01

Message an assistant vs review a branch

In Continuum

Coding day

09:00

Two worktree sessions on the product repo, Claude on the bug and Codex on the tests. Gauges show how hot this week already is.

12:00

Plan approve from the iPhone at lunch. Back at the desk the diff is waiting, not merged.

15:00

Stage three hunks, discard the clever refactor nobody asked for, open the PR from the same pane with checks attached.

18:00

Ledger by repo closes the week. OpenClaw, if installed, is still handling reminders somewhere else entirely.

In OpenClaw

Assistant day

08:00

You message the gateway from WhatsApp on the train. It triages email and drafts the two replies you always write.

12:00

A Slack thread asks it to pull a number and chart it. Canvas on the phone node shows the result without a laptop.

16:00

It runs a scripted repo chore end to end. Useful, and genuinely coding, but there is no worktree and no staged diff between it and main.

21:00

The gateway keeps running on the home server. Continuum is closed with the laptop, which is exactly the difference.

02

Monday path

Monday when the job is shipping code

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

01

Open Continuum’s Code tab for repo work, not the assistant thread. Different tool, different gate.

02

Spawn a Claude or Codex worktree session per ticket so parallel agents never share a working tree.

03

Keep plan mode on for anything near main; read the plan, then approve from the Mac or the paired iPhone.

04

Stage the diff by hunk before you open the PR. The point of the workbench is that you saw the change.

05

Watch the 5h and weekly rails on the menu bar so a Friday batch does not meet a cap you did not know about.

06

If OpenClaw is running, give it its own host, its own model keys, and no write access to repos Continuum ships from.

07

Name Continuum windows by repo or ticket so they never blur into assistant chats on the same machine.

08

Friday: cost by repo in Continuum; assistant spend stays in its own metering.

How Continuum focuses on coding agents

How Continuum focuses on coding agents

  • Sessions are coding-agent first: Claude over PTY, Codex over the ACP harness, plus Cursor, Grok, Gemini, and OpenCode, each in a git worktree.
  • A git checkpoint is taken before every prompt, so a turn that goes wrong is recoverable instead of forensic.
  • The review pane is the product: per-hunk stage and revert, commit sheet, PR status, and merge without leaving the tab.
  • Mobile clients target session control (plan approve, interrupt, glance), not general conversation.
  • The daemon accepts peers on loopback and Tailscale ranges only, and every request carries a bearer token. There is no public chat network in the path.
  • Usage analytics price coding session history locally: dollars by repo, provider, and day, beside live plan windows.
  • Multi-account isolation uses per-instance config dirs so a second Claude or Codex subscription cannot bill the first.
03

Adjacent self-host matrix

Both run on hardware you own, so ownership is not the axis. The axis is shape: OpenClaw optimizes for reaching one assistant from anywhere, Continuum optimizes for turning agent output into a reviewed pull request.

Capability Continuum OpenClaw
Official coding CLI hosts Claude PTY · Codex ACP · Cursor · Grok · OpenCode Bring hosted, gateway, or local models
Reachable from chat appsWhatsApp · Slack · Telegram · Signal Own apps and web only 10+ built-in channels, plus plugins
Runs shell and edits files Full agent session in a worktree Core capability
Git worktree per task Worktree sessions + spawn grid Not a worktree product
Per-hunk diff review Stage, revert, commit in-pane Review in your own tools
PR status + merge PR pane + gh, with checks Via repo tooling it drives
Pre-turn safety checkpoint Git checkpoint before each prompt Allowlists, access groups, mention gating
Live plan quota gauges Menu-bar 5h + weekly rails No plan-window instrument
Multi-account subscription isolation CLAUDE_CONFIG_DIR / CODEX_HOME isolation One credential set per gateway
Local $ by repo Local JSONL → $ by repo/provider/day Your own metering
Voice, camera, canvas surfaces Dictation on Mac, voice reply on Watch iOS and Android nodes
Personal life automations Deliberately out of scope The design center
Open source licence Free app, commercial product MIT, foundation-governed
04

Job boundaries

01 · Shape

Chat thread versus branch.

OpenClaw’s primitive is a conversation you can reach from anywhere. Continuum’s primitives are the worktree, the hunk, and the pull request. Both run agents on your machine; only one is organized around what git will accept.

02 · Reach

Where OpenClaw wins outright.

Ten-plus built-in channels, a plugin system for more, and native macOS, iOS, and Android nodes with Canvas, camera, and voice. Continuum has no chat bridge and is not trying to build one. If reachability is the job, OpenClaw is the better tool and this page will not pretend otherwise.

03 · Review

Continuum’s whole argument is the gate.

Plan mode, a git checkpoint before every prompt, per-hunk staging, and a PR pane with checks. That machinery only makes sense when the output is code someone has to own. See reviewing AI-generated code.

04 · Blast radius

Different networks, different exposure.

OpenClaw’s gateway sits on public messaging networks by design and ships allowlists, access groups, and mention gating to control who can talk to it. Continuum’s daemon accepts only loopback and Tailscale peers with a bearer token. Neither is a verdict, but scope your credentials accordingly: agent security.

05 · Money

Plan windows versus model bills.

Continuum instruments the 5h and weekly windows of the Claude and ChatGPT plans you already pay for, stacks multiple accounts, and prices every session by repo. OpenClaw leaves metering to whichever provider you pointed it at.

05

Coexistence

Stack recipe

How people run both.

Stack recipe: OpenClaw as the always-reachable assistant on its own host and its own least-privilege credentials; Continuum on the workstation where the repos, the git tokens, and the review gate live. They coexist happily on one Mac. The line worth holding is credential scope: an agent reachable from a public messaging network should not hold the keys that can merge to main.

Download Continuum Providers
06

Assistant gateway or coding workbench

Continuum

Choose Continuum when the output is a pull request

  • The work ends in a pull request someone has to review.
  • You want a git worktree per agent so parallel runs cannot collide.
  • You want to stage the diff by hunk before anything reaches main.
  • You run Claude or Codex subscriptions and want their plan windows on screen.
  • You need more than one account isolated cleanly on one machine.
  • You want dollars by repo without standing up a telemetry stack.
OpenClaw

Choose OpenClaw when reachability is the product

  • You want to reach one assistant from WhatsApp, Slack, or Telegram.
  • Life admin, briefings, and automations matter more than PR review.
  • You want it running on a home server long after the laptop closes.
  • Canvas, camera, and voice on phone nodes are part of the appeal.
  • You prefer an MIT, foundation-governed project you can fork.
  • Coding is an errand the assistant runs, not the reason you installed it.
07

What you pay

Continuum

App + your labs

$0 for the app on Mac, web, iPhone, and Watch. Bring your own keys or the Claude and ChatGPT plans you already pay for, and Continuum takes no cut. Optional hosted inference: Plus $25/mo, Max 100 $100/mo, Max 200 $200/mo, Ultra $500/mo.

OpenClaw

Their bill

MIT licensed with no seat price and no first-party hosted plan as of August 2026. Your bill is model tokens plus whatever keeps the gateway online: a home Mac, a VPS, or a one-click template on a platform like Railway. The always-on host is the cost people underestimate, because an assistant that hibernates is not an assistant.

How to compare

Total cost of work

Two budgets, not one. Assistant hosting is a standing monthly cost; coding agent spend rides plans you already own and is what Continuum’s usage dashboard exists to make visible.

OpenClaw plans checked 3 August 2026 against their pricing page. Vendors reprice often - if you spot a stale figure, tell us and we will fix it. Continuum never marks up a session it runs under your own login.

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

08

Questions

Continuum vs OpenClaw.

Deep dives: docs, providers, sessions.

A self-hosted personal AI assistant. One Node.js gateway process runs on your machine or server and bridges chat apps such as WhatsApp, Slack, Telegram, Signal, Discord, and iMessage to an agent that can run shell commands, write code, and use your tools. It began as Clawdbot in November 2025, is MIT licensed, and is now stewarded by the nonprofit OpenClaw Foundation.

Yes. It is MIT licensed with no seat price and no official hosted plan as of August 2026. You pay for model tokens and for the machine that keeps the gateway online.

Yes, and this page does not pretend otherwise: it runs shell commands, edits files, and works with GitHub repos. What it does not ship is the repo-shaped review layer, meaning a worktree per task, per-hunk diff staging, and a PR pane with checks. That is Continuum’s job.

Only if you were using OpenClaw purely for repo work. Continuum is a coding workbench with no chat bridges and no life-automation ambitions; it will not replace an assistant you message from WhatsApp.

Yes. Different processes, different ports, different jobs. The one rule worth keeping is credential scope: give OpenClaw its own model keys and no write access to the repos Continuum ships from.

It is local-first. The daemon runs on your Mac or an enrolled host, transcripts and analytics stay local, and the paired clients reach it over loopback or Tailscale with a bearer token. That is a narrower kind of self-hosting than a general assistant stack, on purpose.

Continuum hosts Claude Code as a first-class engine with worktrees, plan gates, a diff pane, quota rails, and multi-account isolation. OpenClaw can drive coding agents, but it is not positioned as that workbench and does not instrument your plan windows.

No. Continuum stores coding session transcripts and local usage analytics for the providers it instruments. There is no personal memory layer and no attempt to model you between sessions.

Both are MIT-licensed always-on self-hosted agents you message from chat apps. OpenClaw leads on channel breadth and companion apps; Hermes leads on persistent memory and a self-improving skill loop. There is a Continuum vs Hermes Agent page with the same job boundaries drawn.

Further reading

Guide

Claude Code slash commands: reference and custom commands

The built-in commands that matter, the bundled skills hiding among them, and how to add your own in a markdown file.

Read guide → Guide

How to install Claude Code on macOS, Windows, and Linux

The native installer replaced npm as the recommended path. Here is every route, who owns updates on each, and how to fix the common breakages.

Read guide → Guide

Claude Code rate limits: the 5-hour window and weekly caps explained

Why you got cut off, when it comes back, and the settings that buy the most headroom.

Read guide →
Begin

Run your agents
in Continuum.

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

Download for Mac See pricing

vendor-neutral · local-first · multi-device