Claude Cowork
Anthropic's hosted team workspace — shared context, persistent artifacts, and collaborative Claude access without any infrastructure.
Claude Cowork is Anthropic's answer to the question: how does a team use Claude together? Each project gives every member shared context — the same system prompt, the same knowledge documents, the same artifact library — so conversations build on each other instead of starting from scratch. No API keys, no server, no DevOps. If your team is non-technical and needs AI-assisted collaboration today, Cowork is the fastest path.
- Apr 17 — Claude Design launched. Anthropic Labs' text-to-prototype tool inside Cowork. Turns prompts and codebases into design systems, websites, and slide decks. Powered by Opus 4.7. See our Claude Design guide.
- Apr 16 — Opus 4.7 + xhigh effort. New flagship model with an
xhigheffort level betweenhighandmax. Use/effortto tune speed vs intelligence. Auto mode is now available for Max subscribers using Opus 4.7. - Apr 9 — Cowork hit GA. General availability with Managed Agents public beta and enterprise RBAC. Business and Enterprise tiers stabilized; SOC 2 Type II current.
- March — 90-day artifact retention. Shared artifacts persist for 90 days from last modification (was 30). Business: pin for indefinite. Enterprise: custom retention periods.
Guides
/effort xhigh vs high vs max, why Claude self-corrects mid-sentence, the wellness-suggestion behavior, and the most common usage-limit gotchas. Updated weekly.Cowork is locked to Anthropic models and a web UI. For model flexibility (swap Claude, Gemini, or local Ollama), messaging-app channels (Telegram, WhatsApp, Discord), or 50–70% lower costs at equivalent usage, see OpenClaw. For long-running autonomous tasks that run unattended, see Hermes. Many teams run Cowork for team documents and OpenClaw for personal automation simultaneously — they don't conflict.
At a Glance
| Factor | Detail |
|---|---|
| What it is | Anthropic's hosted team collaboration product for Claude |
| Who runs the infrastructure | Anthropic — no servers, no API keys, no config files |
| Model access | Claude Haiku 4.5, Sonnet 4.6, Opus 4.6, and Opus 4.7 with xhigh effort tier (tier-dependent) |
| Effort levels | low · medium · high · xhigh · max — set via /effort or model picker (Opus 4.7) |
| Managed Agents | Public beta — included on Business and Enterprise during the beta window |
| Claude Design | Available since April 17, 2026 — text-to-prototype on Pro and above |
| Model locked to Anthropic | Yes |
| Pricing model | Per-user/month subscription (not per-token) |
| Tiers | Free · Pro (~$20/user/mo) · Business (~$30/user/mo) · Enterprise (custom) |
| Shared context | Yes — Projects with system prompts, knowledge documents, and artifact libraries |
| Artifact retention | 90 days from last modification (updated March 2026); Business: pin for indefinite |
| Official integrations | Google Drive, GitHub, Slack, Jira, web search, code execution |
| SSO | SAML/OIDC on Business and Enterprise; SCIM on Enterprise |
| SOC 2 compliance | Business and Enterprise tiers |
| Data residency | Enterprise only (US or EU) |
| Time to first useful output | Minutes — create workspace, invite team, start a project |
| Can embed in customer-facing products | No — Cowork ToS requires internal team use only; use the Claude API for that |
Cowork & Claude Code Troubleshooting
- Rate limit 429 — what to do when you hit per-minute caps
- MCP server not responding — diagnose .mcp.json + claude mcp logs
- All troubleshooting entries →
Cowork Use Cases & Security
- Code review automation — pairs naturally with Cowork's GitHub integration
- Release notes generator — Projects keep audience context across runs
- Social content engine — leverages the new Claude Design slide-deck mode
- Don't paste secrets into Cowork chats — retention defaults make this risky
- 15-minute hardening checklist for any Cowork workspace
See also: OpenClaw · Hermes · IronClaw · Cowork vs Claude API vs OpenClaw · Decision guide