Manus AI is the autonomous general-purpose agent from Butterfly Effect Pte. Ltd. (Chinese-founded, now Singapore-headquartered) that went viral in March 2025. CRHQ is a managed multi-agent operations platform on a dedicated VPS per customer. Both are positioned as "autonomous AI that does work for you" — but the architectures, persistence models, and trust assumptions are very different.
At a glance — what each platform ships out of the box
TL;DR
| Manus | CRHQ | |
|---|---|---|
| What it is | Shared cloud agent platform | Managed dedicated VPS per customer |
| Underlying models | Claude 3.5/3.7 Sonnet + fine-tuned Qwen | BYO — Anthropic, OpenAI, Bedrock, Vertex |
| Compute model | E2B Firecracker microVM per task (disposable) | Persistent VPS with Postgres, file system, browser |
| Persistence | Session event-stream + files, VMs are disposable | Postgres + project documents + per-agent memory |
| Self-host | Not offered | n/a — managed by CRHQ on your dedicated VPS |
| Data residency | Opaque, Chinese-origin platform (banned by some US states) | Dedicated VPS, your provider region |
| Scheduling | Limited | First-class scheduling per recipe |
| Pricing | Free / $20 / $40 / $200 / Team ~$39/seat | $299–$399/mo per satellite + BYO model |
What Manus actually is
Manus runs an analyze-plan-execute-observe loop using the CodeAct pattern (executes Python instead of JSON tool calls). Each task spins up a dedicated Ubuntu microVM (E2B Firecracker) with shell, real Chromium, file system, and Python/Node — visible to the user as a "virtual computer in the browser" you can watch the agent operate.
The intelligence is not proprietary — Manus orchestrates Claude 3.5/3.7 Sonnet for primary reasoning and fine-tuned Qwen for sub-tasks. The product is the orchestration, the VM, the browser, and the UX.
Tier structure:
- Free — 300 credits/day, 1 task at a time
- Standard $20, Pro $40, Extended $200
- Team ~$39/seat, 5-seat minimum, shared credit pool, SSO
There is no self-hosted option. Community clones (OpenManus, AgenticSeek, Suna) fill that gap.
Reception has been "capable but unreliable." MIT Tech Review called it "an intelligent intern" but flagged crashes, captchas, paywalls, and getting stuck on long tasks. Persistence is weak — VMs are disposable; memory lives in the session event-stream and externalized files. The optional Cloud Computer add-on provides a 24/7 VM, but that's a paid extra rather than the default.
Warning: Data residency matters. Manus is a Chinese-origin platform now headquartered in Singapore. It has been banned by Tennessee state government, and in April 2026 China's NDRC blocked Meta's $2B acquisition over data-flow risk. If your work involves customer data, regulated data, or competitive information, factor this in.
What CRHQ is
CRHQ provisions a dedicated managed VPS (a satellite) per customer with Postgres, file system, browser, persistent memory, encrypted credentials, and a multi-agent team. You enable skills, give plain-English instructions, and optionally schedule them to run on a cadence. The VPS is yours — same provider, same region, every time. Memory persists. Files persist. Browser login state persists.
Where Manus sells a shared SaaS where you watch a fresh VM do your task, CRHQ sells your own dedicated server where a team of agents builds up institutional knowledge over weeks and months.
Side-by-side
| Capability | Manus | CRHQ |
|---|---|---|
| Compute model | Disposable Firecracker microVM per task | Persistent dedicated VPS per customer |
| Persistence across tasks | Weak — session event-stream + files | Strong — Postgres + project documents + per-agent memory |
| Browser session continuity | New Chromium per task; state via files | Persistent Chromium with login state, anti-bot tuning |
| Multi-agent team | Single Manus agent (orchestration internal) | Researcher, Developer, Writer, Analyst, Monitor + delegation skill |
| Skills / extensibility | Limited public skill model | ~25 versioned platform skills, toggleable per satellite |
| Slack / Fireflies / GA4 / Sentry | Generic via browser/Python | Built-in skills with proper APIs |
| Scheduling | Limited | First-class per recipe |
| Self-host | No | n/a — CRHQ manages on your dedicated VPS |
| Data residency / region control | Opaque | Pick your VPS provider/region |
| Audit log | Per-user, on Manus servers | On your satellite, exportable |
| Watch-the-VM UX | Yes — distinctive feature | Optional via screenshots; primary UX is web app |
| Model freedom | Internal — Claude + Qwen | BYO — Anthropic, OpenAI, Bedrock, Vertex |
Where Manus wins
- The "watch the VM" UX is genuinely impressive. Seeing a real Chromium and a real terminal execute your task in real time is a great demo and a great debugging surface for one-off work.
- Zero ops setup. Sign in, type a task, watch it go. No infrastructure to think about.
- Free tier is generous for evaluation (300 credits/day).
- Cheapest paid entry is $20.
- Strong on one-off research/data tasks that fit inside a single session.
If you want a powerful general-purpose agent to play with, kick off occasional tasks, and aren't sensitive to data residency or persistence, Manus is fun.
Where CRHQ wins
- Persistence. Project documents, per-agent memory, browser login state, and file system all survive across sessions and across days. Manus VMs are disposable by design.
- Multi-agent fleet with explicit roles. Researcher, Developer, Writer, Analyst, Monitor — pre-built and able to delegate to each other. Manus is one orchestrator.
- A real skills system. Slack, Fireflies, Sentry, GA4, Search Console, Stripe, Postgres, agent-browser — versioned, tested, and toggleable. Manus equivalents are "tell the VM to install something."
- Recurring jobs as a primitive. "Every Monday at 8am" is a single scheduled recipe. In Manus, recurring work needs the Cloud Computer add-on plus your own logic.
- Dedicated VPS, your region, your control. Important for regulated industries, EU residency, customer data, and any work where "where does the data go" has an answer that matters.
- Model freedom. BYO Anthropic, OpenAI, Bedrock, Vertex.
- Audit + backup baseline designed for business operations rather than consumer-style usage.
Can I replicate CRHQ on top of Manus?
Mostly no. The architectures point in opposite directions: Manus is "fresh VM per task, results out, VM gone"; CRHQ is "persistent server with accumulated state." You could chain Manus tasks together and externalize state to your own database, but you're then doing the heavy lifting of CRHQ on top of a product that wasn't designed for it. And you still don't get the dedicated VPS, the multi-agent team, the integration skills, or the data residency control.
Pricing
| Manus | CRHQ | |
|---|---|---|
| Free | 300 credits/day, 1 task | n/a |
| Entry | Standard $20/mo | $299/mo (Standard satellite) |
| Mid | Pro $40/mo | $349/mo (Pro satellite) |
| Heavy | Extended $200/mo | $399/mo (Premium) |
| Annual discount | Varies | 20% off |
| Model usage | Included | BYO API key (Anthropic / OpenAI / Bedrock / Vertex) |
Manus is cheaper at the entry tier for one-off use. CRHQ amortizes faster the moment you need persistence, scheduling, or regulated-data assurances.
Bottom line
- Pick Manus if you want a powerful general-purpose agent for occasional one-off tasks, you love the "watch the VM" UX, and data residency / persistence aren't sticking points.
- Pick CRHQ if you want a persistent multi-agent operations platform on a dedicated VPS you control, accessed by a team, with a real skills library and scheduling baked in — and you want to know exactly where your data lives.
Both are managed services. Only one gives you a server you can point at.