Anthropic Orbit is a proactive AI assistant that ships inside Claude Cowork and pulls personalized insights from Gmail, Slack, GitHub, Calendar, Drive, Figma, and other workspace apps without waiting for you to ask. The product flips the dominant assistant pattern — reactive question-and-answer — into something closer to a collaborator who already read the morning’s traffic and walks into your day with context. For knowledge workers drowning in twelve dashboards and fourteen unread channels, that’s a meaningful UX shift, and the architecture choices Anthropic made under the hood signal where the assistant category is heading next. This is the practical guide: what’s actually shipping, what it means for your workflow, and how to evaluate whether to adopt now or wait for the next iteration.
What’s actually new
Three things distinguish Anthropic Orbit from the existing crop of Claude-based productivity tools and from competing offerings (Microsoft Copilot, Google Duet, OpenAI’s ChatGPT enterprise integrations).
First, genuine proactivity. Orbit doesn’t sit and wait for prompts. It runs continuous, low-cost summarization passes over the integrated workspaces, identifies what would genuinely change your behavior in the next few hours (a meeting got rescheduled, a critical PR review request landed, a customer thread escalated overnight), and surfaces those at the right moment. The proactive layer uses small, fast model variants for triage, escalating to the larger Claude family only when something requires real reasoning. The cost economics make continuous monitoring viable.
Second, typed connector architecture. Orbit’s integrations aren’t generic OAuth scopes; they’re typed, scoped MCP-style connectors with explicit data lineage. When Orbit synthesizes a morning briefing, you can drill into any insight and see “this came from message X in Slack channel Y, sent by Z at time T, combined with calendar event N.” The auditability is unusual for an assistant product and is one of the things that lets Anthropic claim enterprise-grade privacy posture without hand-waving.
Third, cross-app reasoning. The killer feature is connecting facts across silos. A typical Orbit insight is “the customer who emailed you Tuesday about the invoice issue is the same one who filed the GitHub bug report this morning, and they’re scheduled in your calendar for 2pm.” Each individual signal is in a different app, none of them flag the connection, and a human juggling tabs misses it half the time. Orbit doesn’t.
The model layering matters. The proactive watch loop uses a Sonnet-class model for triage (cheap, fast, runs continuously). The synthesis layer that builds your morning briefing or alerts you to cross-app patterns uses Opus 4.6 / 4.7 (smarter, slower, runs on demand). Reasoning-heavy tasks — investigating a complex incident, drafting a multi-stakeholder reply — escalate to the Opus variant with explicit user invocation. The result: continuous monitoring is affordable, deep reasoning is available when needed.
Why it matters
- The reactive-assistant ceiling is real. Reactive tools improve as long as you remember to ask them. Most knowledge work fails at the “remember to ask” step. Proactive tools that surface what matters without prompting unlock value the reactive paradigm structurally cannot reach.
- Cross-app context is the dominant unsolved problem. Every individual SaaS adds another silo. The companies that lose to information overload aren’t the ones with bad tools — they’re the ones whose tools don’t talk to each other. Orbit’s typed connectors are a credible attempt to fix this.
- Privacy posture is a differentiator now, not a tax. Microsoft Copilot has had buyer hesitation tied to data-handling concerns. Orbit’s typed-data-lineage approach gives security and compliance teams something concrete to evaluate. Enterprises that ruled out earlier proactive assistants on privacy grounds may reconsider.
- The model-tier layering is the architecture pattern of 2026. Continuous monitoring with small/fast models, on-demand reasoning with frontier models — this pattern will define every successful productivity AI product going forward. Orbit is an early production deployment of it; expect imitators within two quarters.
- Calendar awareness changes the assistant relationship. When the AI knows what meeting you’re walking into, suggestions get sharply more useful. “Brief me before my 2pm” is a different kind of prompt than “summarize this thread.” Orbit makes the former cheap.
- Adoption friction matters more than capability. Orbit’s success will turn on how easily it integrates into existing workflows, not on how clever its synthesis is. Rough integration UX kills proactive tools faster than any other failure mode.
How to use it today
Anthropic Orbit is rolling out through Claude Cowork’s enterprise tier, with a phased availability schedule across organizations. If you have access — or want to evaluate before deciding — the practical setup runs in roughly 20 minutes.
- Confirm Cowork eligibility. Orbit ships only in Claude Cowork (the team / enterprise tier), not the consumer Claude.ai. Sign in to console.anthropic.com → check your workspace tier. If you’re on Claude Pro individual, you’ll need to upgrade or request a team trial.
- Connect the integrations one at a time. The Orbit setup wizard walks through each: Gmail, Slack, GitHub, Calendar, Drive, Figma, and the long-tail integrations (Linear, Notion, Asana, Jira). Each is a standard OAuth flow with explicit scope review. Don’t connect everything at once on day one — pick the two or three apps where most of your work happens and add others as you learn what kinds of insights are useful.
# Once connected, validate access via the Orbit health-check API curl -X GET https://api.anthropic.com/v1/orbit/health \ -H "Authorization: Bearer $ANTHROPIC_KEY" \ -H "anthropic-version: 2026-05-01" # Expected response shape: # { # "integrations": [ # {"id": "gmail", "status": "connected", "last_sync": "2026-05-05T14:23:00Z"}, # {"id": "slack", "status": "connected", "last_sync": "2026-05-05T14:23:00Z"}, # ... # ] # } - Tune the proactive cadence. Orbit’s default is “morning briefing + real-time alerts for high-priority signals.” Most users want to dial down the alert volume in the first week. Settings → Orbit → Notification Cadence → set urgent-only for the first seven days, expand once you’ve calibrated to your noise tolerance.
- Configure the privacy boundaries. Orbit’s connector model lets you scope per-integration. For Slack, exclude specific channels (HR, exec-only) from the index. For Gmail, exclude labels (personal, financial). For Drive, restrict to specific shared drives. Doing this up-front beats discovering Orbit summarized something it shouldn’t have.
- Use the API for custom workflows. Orbit exposes a REST API for programmatic use. The most common patterns: pulling the morning briefing into a Slackbot, generating a daily email digest, or wiring Orbit insights into your existing dashboard.
# Fetch today's morning briefing programmatically curl -X POST https://api.anthropic.com/v1/orbit/briefing \ -H "Authorization: Bearer $ANTHROPIC_KEY" \ -H "anthropic-version: 2026-05-01" \ -H "Content-Type: application/json" \ -d '{ "user_id": "u_abc123", "context_window": "today", "format": "markdown", "max_items": 8 }' # Streaming variant for real-time alerts: curl -X POST https://api.anthropic.com/v1/orbit/stream \ -H "Authorization: Bearer $ANTHROPIC_KEY" \ -H "anthropic-version: 2026-05-01" \ -d '{"priority_threshold": "high"}' \ --no-buffer - Establish review hygiene. Once a week, scan Orbit’s surfaced insights and rate which were genuinely useful vs noise. The product learns from feedback, but only if you give it. Five minutes of weekly tuning produces meaningfully better insights by week three.
- Watch your team’s adoption signals. If you’re rolling Orbit out to a team, track three things weekly: percentage of team members opening the morning briefing, percentage acting on at least one Orbit-surfaced item, and explicit “this insight was useful / not useful” feedback. The adoption curve looks healthy when all three trend up over the first month.
How it compares
| Product | Provider | Proactivity | Cross-app reasoning | Privacy posture | Pricing (per user / mo) |
|---|---|---|---|---|---|
| Anthropic Orbit | Anthropic | Continuous + on-demand | Native, typed connectors | Per-integration scoping, lineage tracking | Bundled with Cowork ($30 base, +$10 Orbit) |
| Microsoft Copilot for M365 | Microsoft | Reactive + scheduled summaries | Within M365 stack only | Tenant isolation; less granular scoping | $30 |
| Google Duet AI | Reactive primarily | Within Workspace stack only | Org-level controls | $30 | |
| ChatGPT Enterprise | OpenAI | Reactive | Via custom GPTs + connectors | SOC 2 + workspace boundary | $60+ (variable) |
| Glean AI | Glean | Reactive search + summarization | Strong, with knowledge graph | Granular, per-source | $15-30 depending on tier |
The clearest differentiator: proactivity. Microsoft Copilot, Google Duet, and ChatGPT Enterprise are all reactive at their core — you ask, they answer. Orbit and Glean both push insights without prompting, but Orbit’s cross-app reasoning is more developed and its privacy story is more granular. Glean is the strongest direct comparison; the practical difference is that Glean is search-and-summarization-focused while Orbit is action-and-decision-focused.
The pricing math is also interesting. At $40/user/month total (Cowork + Orbit), Anthropic is positioning Orbit as a productivity-suite competitor, not a pure assistant. Microsoft Copilot is the obvious price reference at $30/user/month — but Copilot is integrated only with Microsoft’s own apps, while Orbit reaches across the SaaS stack most enterprises actually use. For a team running Slack + GitHub + Notion + Figma alongside or instead of M365, Orbit covers more surface area for marginal additional cost.
What’s next
Three trajectories worth watching for the rest of 2026.
First, connector breadth will expand fast. Anthropic shipped Orbit with seven core integrations and is publicly committed to thirty by end of year. The interesting question is which long-tail integrations get prioritized — vertical-specific tools (Salesforce, HubSpot, Workday for HR-heavy orgs) versus developer-stack additions (Sentry, PagerDuty, Datadog for engineering teams). The pattern of which integrations land in which order will tell you which buyer segment Anthropic is courting hardest.
Second, the proactive-assistant pattern will get copied. Microsoft, Google, and OpenAI all have the model capability to ship something equivalent. The Microsoft variant is most likely to land first given the integration depth Microsoft already has across M365. Google has the data but historically slow product velocity in this category. OpenAI ships product fast but lacks Anthropic’s typed-connector architecture story. Expect at least one major-vendor announcement of a comparable product in the next two quarters.
Third, regulation and compliance frameworks will catch up to proactive AI. Today, most workplace-AI compliance frameworks were written for reactive tools — they assume the AI does what you ask, not that it surveys your work continuously. Proactive assistants raise different questions: what data is the AI looking at, when, with what retention, and under whose authority. Expect updated SOC 2 control points, new ISO controls, and possibly explicit regulatory guidance in regulated industries before year-end.
The deeper shift Orbit signals is the move from “assistant as a tool” to “assistant as an environment.” A reactive tool sits in your tab list; you visit it when needed. A proactive environment surrounds your work; it notices patterns you miss and changes what you pay attention to. That’s a different relationship — closer to a colleague than a search box. Whether Orbit specifically wins this market is uncertain. That this market exists, and is the next big one, is no longer in doubt.
Frequently Asked Questions
Is Orbit available on the consumer Claude tier?
No. Orbit is a Claude Cowork (team / enterprise) feature only. Individual Claude Pro accounts do not have access. The architectural reason: Orbit’s value depends on connected workspace apps that mostly aren’t relevant to a single-user consumer flow. Anthropic has not announced consumer plans.
What workspace apps integrate with Orbit at launch?
Gmail, Slack, GitHub, Google Calendar, Google Drive, Figma, plus initial support for Linear, Notion, Asana, and Jira. Anthropic has committed publicly to expanding to thirty connectors by end of 2026, with Salesforce, HubSpot, Microsoft 365, and Confluence as the most-requested next adds.
How does Orbit handle privacy and data residency?
Per-integration scoping (you can exclude specific channels, labels, drives), data-lineage tracking on every surfaced insight, regional data-residency options (US and EU as of launch, more regions on roadmap), and standard enterprise security controls (SOC 2, ISO 27001 in progress, HIPAA BAA available on request). Anthropic publishes its data-handling policies in detail at trust.anthropic.com.
What’s the pricing relative to Microsoft Copilot or Google Duet?
Roughly $40/user/month (Cowork base $30 + Orbit add-on $10), versus Microsoft Copilot for M365 at $30/user/month and Google Duet at $30/user/month. Orbit is positioned as a premium offering with broader integration surface; the value depends on how much of your workflow lives outside the Microsoft or Google ecosystems.
Can I use the Orbit API to build custom integrations?
Yes. The Orbit API exposes endpoints for fetching briefings, streaming alerts, querying insights, and submitting feedback. Most teams use it to pipe Orbit insights into Slack, email digests, or internal dashboards. API documentation is at docs.anthropic.com/orbit.
Should we adopt Orbit now or wait for competitors to catch up?
Depends on your stack. If your team relies on a heterogeneous SaaS mix (Slack + GitHub + Notion + Figma + Drive), Orbit covers more surface area than any single-vendor alternative today. If you’re predominantly Microsoft 365, Copilot’s deeper integration there may serve you better. If you can run a 30-day pilot with five users, do it — the value of proactivity is hard to evaluate without hands-on time, and a structured pilot tells you in a month what speculation can’t.