Reclaim AI

Writes into external systems happen under OAuth, and task systems sync both ways. Nine named integrations each have their own product page: Slack, Zoom, Google Tasks, Todoist, Asana, Jira, ClickUp, Linear and Raycast, plus Google Calendar and Microsoft Outlook as the core calendar surfaces.

Authenticated writes are documented in the permission scopes themselves. Google: "view and edit events on all your calendars", used to "create, update, and remove tasks, habits, buffer time, and synced events from your calendars". Microsoft Graph: Calendars.ReadWrite, which "allows the app to create, read, update, and delete events of all calendars without a signed-in user". Google Tasks: "create, edit, organize, and delete all your tasks".

Slack: chat:write, users.profile:write and dnd:write, used to post messages and change the user's status and Do Not Disturb state. Gmail: create new drafts. Writing into five external systems under OAuth goes well beyond retrieval. Task systems sync two ways: completion status is written back to the source so the two stay consistent. Further integrations are marked coming soon.

Sourcereclaim.ai/security permission scopes section and reclaim.ai/integrationsread 2026-09-12

Reclaim runs a permanent multi-constraint optimization loop that executes unattended. The published feature set is the constraint set: AI Focus Time with weekly goals, AI Habits with recurring windows, AI Tasks with priority tiers and time estimates, AI Smart Meetings solving across all attendees' schedules, AI Buffer Time for breaks and travel, AI Calendar Sync across multiple calendars, Working Hours, Team OOO Calendar and No Meeting Days. Each has its own product page, and all are solved together.

Execution is real and unattended, shown by the write scopes rather than by marketing: Reclaim creates, updates and removes events on the user's calendars, and the Microsoft scope is explicit that it acts "without a signed-in user". The agent displaces lower-priority work and reschedules it when a higher-priority item needs the slot, reacting to any calendar change rather than on a batch. No named runtime or BPMN is published; the work is continuous multi-step execution against interacting constraints.

Sourcereclaim.ai/security and reclaim.ai product navigationread 2026-09-12

A rich structured state model grounds the agent, and nothing else does; there is no document corpus. The agent works from task due dates, priority tiers and time estimates imported live from six external systems, habit definitions and flexibility windows, weekly focus goals, attendee availability and timezones, declared travel periods, working hours, and accumulated time-tracking history.

The vendor says it "stores data from the primary calendar associated with your Reclaim account" because that "allows us to do things like reschedule your tasks and habits when conflicts occur, use prediction services to flip events between free and busy, and run processing against your events to automatically categorize them". That is a persistent, queried state model.

There are no notes, docs or wiki, no retrieval or semantic search over content, no embedding layer and no RAG surface. Reclaim's own data minimization makes this a design choice: it does not store data from any calendar other than the primary one, holding sync data only in memory and discarding it when the job completes.

Sourcereclaim.ai/security Data Usage sectionread 2026-09-12

The agent is bounded before it runs and everything is shown afterward, but no gate sits in between. The user sets everything that constrains the agent in advance: priority tiers determining what may be displaced, habit windows fixing when routines land, weekly focus goals, no-meeting days, buffer rules, working hours and travel periods.

Permissions are requested just-in-time, and the vendor says so: "We ask for these permissions just-in-time, meaning that we only ask for permissions when they are absolutely necessary", adding that "we never ask for more permissions than we need". Optional integrations are opt-in, with the Slack, Google Tasks and Gmail scopes each marked as needed "only if you choose to install and use" that integration. Access is revocable at the identity provider, and self-service deletion wipes calendar data "within one hour" and removes Reclaim-created events. Every action the agent takes is visible on the calendar and reversible in one drag.

The agent moves real events without asking. There is no approval gate, preview before commit, policy engine, per-action risk calibration or administrator surface configuring any of it. Oversight sits before the run and after it, never during, and that is the design rather than an omission.

Sourcereclaim.ai/security permissions and data deletion sectionsread 2026-09-12

A strong, verifiable attestation set sits beside identity controls the customer runs. For customer access, Reclaim offers "native SSO and SCIM integrations with Okta and OneLogin, as well as extensibility for any OIDC platform". That gives identity federation plus automated provisioning and deprovisioning. The security page's least privilege and 2FA section covers Reclaim's own employees, alongside background checks, security awareness training and policy acknowledgments; those are internal controls.

On attestation, Reclaim holds SOC 2 Type II with zero exceptions, audited by A-Lign. GDPR, CCPA, and Data Privacy Framework certification "independently verifiable on dataprivacyframework.gov", which a buyer can confirm on a government register without contacting the vendor. A SafeBase trust center at trust.reclaim.ai, a published subprocessors page, a DPA at reclaim.ai/dpa incorporated into the Business Terms, a responsible disclosure policy, AES-256 at rest with rotating keys, forced TLS, external OWASP and SANS scanning, static code analysis and annual BC/DR testing.

The trust center also discloses an open penetration test finding by number, explains it was mitigated by WAF rate limiting, and says retesting is impractical without denial-of-servicing itself. Disclosing what is not fixed is unusual, and to the vendor's credit.

Sourcereclaim.ai/security, trust.reclaim.ai and reclaim.ai/blog/clockwise-shutdown-newsread 2026-09-12

Every decision is visible, but no record is kept of why a decision was made. Visibility comes by construction: the agent's output is the calendar, so every scheduling decision is displayed by default, inspectable at a glance and reversible immediately. AI Time Tracking analyzes where hours actually went, Workforce Analytics extends that across a team, and schedule and workload insights surface in Slack.

Seeing what happened is not the same as reconstructing why. Nothing published records which constraint won when an event moved. There is no customer-facing audit log, decision trail, attribution, retention policy for agent activity or export. A user sees that a task was displaced and cannot learn whether it lost to a deadline, a priority tier, an attendee's availability or a focus goal.

The security page's "Monitoring & observability" section (infrastructure and application performance monitoring, error handling, "centralized application and infrastructure logging") is Reclaim watching its own estate, invisible to the buyer. Audit logging appears in the trust center only as a security control category, not as a surface a buyer can use.

Sourcereclaim.ai/security Software Development & Operations section and reclaim.ai product navigationread 2026-09-12

Configuration and the application's own data persist; no agent memory is documented. Habits recur indefinitely, priority assignments and flexibility windows persist, travel and no-meeting-day settings hold, time tracking accumulates, and the scheduler carries plan state across sessions. Every one of those is the application's own data model or persistent configuration, not a memory layer. There is no per-user memory scope distinct from the account, no published lifetime, no expiry or purge path for memory as opposed to the account, and no read or write surface.

Nothing published says the system learns a user's preferences over time or keeps an adaptive model of individual behavior, and the Slack conversational surface has no documented cross-session recall. Reclaim 2.0's Schedule Assistant is a conversational surface, and no memory is documented for it either.

Sourcereclaim.ai/security and reclaim.ai product navigationread 2026-09-12

Reclaim hosts all data in the United States and offers no residency choice. The security page states the fact and the limit in one place: "All data is hosted in the USA, with redundancy between us-east-1 and us-east-2 regions." Two regions inside one country for failover is availability engineering, not customer deployment control. No region selection, EU residency option, on-premises or air-gapped mode, private VPC or single-tenant tier is published.

Reclaim's legal transfer position is strong: Data Privacy Framework certification verifiable on a government register, standard contractual clauses through a published DPA, and GDPR and CCPA commitments. A lawful basis for moving data to the United States is not data residency, though, and a European buyer with a localization obligation is not served, whatever the paperwork says.

Sourcereclaim.ai/security Data & Infrastructure Hosting sectionread 2026-09-12

Prebuilt scheduling behaviors are ready for a user to turn on rather than build, and an AI Agent Library is published beside them. Habits, Smart Meetings, Scheduling Links, Focus Time, Buffer Time, No Meeting Days, Calendar Sync, Working Hours, Time Tracking and Team OOO Calendar each have their own product page.

The AI Agent Library is a named product surface at reclaim.ai/ai-agent-library, listed in the Products column of the site footer, with at least one addressable sub-page at /ai-agent-library/no-meeting-days; the footer maps its Use Cases entry to the same path. Six named Initiatives are separately addressable under /initiatives/: Focus Time, Calendar Fragmentation, Meeting Agenda, Buffer Time, Slack Notifications and No-Meeting Day. A library and a set of addressable units exist, but nothing published shows whether they are adoptable working agents or marketing program pages.

Sourcereclaim.ai/security footer navigationread 2026-09-12

Work reaches Reclaim without anyone invoking it, but delivery stays on the employee's own work surfaces. Calendar events arriving or moving fire immediate re-optimization: the Google scope exists partly to "respond to RSVP signals" and to be "notified when those settings change". Task creation, due-date changes and priority changes in six external systems propagate in, habits run on recurring schedules, and buffer rules fire on meeting boundaries. Reacting to state changes across six third-party systems is a wide event surface.

Delivery is confined to surfaces the employee already occupies: the calendar itself, Slack, Raycast, the Gmail and Google Calendar add-ons and the web app, with no native mobile client. Outward reach is limited to Scheduling Links for external booking and webhooks to a customer endpoint.

There is no customer-facing channel, no messaging surface to third parties and no voice or web surface, because the product faces the employee rather than a market.

In Slack, the mcp:connect scope powers a conversational Schedule Assistant so a user can "ask the assistant to find time, reschedule, protect focus, or resolve conflicts directly inside Slack", an invocation surface rather than outbound reach.

Sourcereclaim.ai/security Slack scopes section and reclaim.ai/integrationsread 2026-09-12

Reclaim names one model provider and offers no choice. Its Slack Marketplace security disclosure states that its use of OpenAI includes OpenAI's Zero Data Retention policy, so no customer data is stored by OpenAI in connection with the service.

Naming the provider and documenting a zero-retention arrangement is a better procurement answer than silence, but the customer has no say: there is one provider, no selection surface, no routing between models, no bring-your-own-model or bring-your-own-key, and no admin model policy.

Reclaim's security page covers hosting, encryption, employee security, policies, vulnerability management, access management, testing, monitoring and per-integration permission scopes in detail, and names no model provider; the disclosure lives on the Slack listing.

SourceReclaim Slack Marketplace security disclosure, against a full read of reclaim.ai/securityread 2026-09-12

The outbound surface is real but narrow: documented webhooks, with no published API reference and no official SDK. Webhooks fire on Scheduling Link bookings ("Reclaim will post to the endpoint when meetings are booked, rescheduled, or canceled") with signed payloads, a rotatable secret, custom data passthrough of up to five parameters, an https-only requirement, no delivery to localhost or IP addresses, and a stated 10-second idempotent-receiver contract. That is a properly specified integration contract. Webhook support is available on the Business and Enterprise plans.

An API exists but is not documented. Keys are self-generated in Settings and third-party automation platforms authenticate with them, but no reference, endpoint list, auth scheme or schema is published on any Reclaim host, and no developer subdomain exists; the help center is the only documentation surface. Developers have built around the gap: a community Python SDK states that it is "not officially affiliated with Reclaim.ai and has been reverse-engineered from the Reclaim.ai web app", working from a Swagger spec passed on informally by an employee.

The Slack app's mcp:connect scope connects the user's Slack workspace to Reclaim's Schedule Assistant. That is Reclaim reaching into Slack to power its own assistant, not Reclaim exposed to outside callers.

Sourcereclaim.ai/integrations/webhooks and reclaim.ai/securityread 2026-09-12

Output is inspected continuously as a side effect of use and performance is measured afterward, but no change can be evaluated before it takes effect. The agent's reasoning output is the primary interface, so every decision is exposed by default, checkable against what the user expected, and correctable in one drag. AI Time Tracking then measures whether it worked, analyzing where hours actually went against the focus goals the user set, and Workforce Analytics extends that to a team. That is retrospective measurement of the agent's own performance rather than of a business outcome.

There is no sandbox or dry run to test a priority scheme or habit configuration before it starts moving real meetings, no evaluation harness, no scored test cases, no regression check when a configuration is edited, and no A/B comparison of scheduling strategies. The security page's "continuous testing" (peer reviews, unit and integration tests, and validation in staging and preview builds) is Reclaim's own software development lifecycle, not a surface the customer can use to test their own configuration.

Sourcereclaim.ai/security and reclaim.ai product navigationread 2026-09-12

Every system Reclaim touches is reached through documented APIs authenticated with OAuth, so there is no browser or computer control. No hosted or local browser, desktop session, or remote or local computer control is documented. The security page enumerates the APIs by name: the Google Calendar API, the Google Tasks API, the Microsoft Graph API with named scopes, and the Slack API with fourteen named scopes. Reclaim publishes its Google OAuth client ID and states that "Reclaim uses Google APIs and is not a marketplace app". Outbound it uses signed webhooks.

A browser extension and Gmail and Google Calendar add-ons exist, but those are Reclaim's own surfaces installed in someone else's interface, not the agent operating a screen. There is no native mobile client, which underlines that this is a web and API product rather than one that drives interfaces.

Sourcereclaim.ai/security Want More Details section and reclaim.ai/integrationsread 2026-09-12