Temporal Cortex

Agents get 18 scheduling tools over Google Calendar and Microsoft Outlook (OAuth 2.0) and CalDAV calendars such as iCloud and Fastmail, including contact search across Google People and Microsoft Graph and a book_slot tool that writes events, reached through the vendor's MCP server, A2A or the REST API. Those are authenticated read and write actions in the customer's calendars.

Sourcetemporal-cortex.com/llms-full.txtread 2026-09-22

A multi-step booking is carried out by Temporal Cortex itself: book_slot locks the slot, verifies there is no conflict, writes the event and releases the lock, rolling everything back if a step fails, with distributed locking coordinating concurrent agents on the managed platform.

The customer's agent sequences the other tools, following a documented orient, resolve, query and book pattern, inside its own framework. That is a vendor run transaction with rollback that is safe to retry; no workflows that combine deterministic nodes with autonomous agent steps, with branching and routing, are documented.

Sourcetemporal-cortex.com/llms-full.txtread 2026-09-22

There is no retrieval structure over the customer's documents or knowledge. Contact search queries Google People and Microsoft Graph live, and the availability tools read calendars live; both are live integration rather than grounding. No ingestion, indexing, retrieval or RAG layer over company data is documented.

Sourcetemporal-cortex.com/llms-full.txtread 2026-09-22

A person can be put in front of an agent's booking: when another agent sends request_booking, the recipient's policies decide whether it is auto-accepted, held for the recipient's approval or blocked, and the API's booking rules carry a confirm_before_booking setting beside minimum notice and advance-booking limits.

Policies set per caller domain or trust level allow or deny requests by priority, and the content firewall strips injection attempts before any tool runs. So an approval step can sit before an agent's booking commits, inserted at the policy layer; how the approval request reaches the person is not described.

Sourcetemporal-cortex.com/llms-full.txtread 2026-09-22

Access controls run deep: SSO through SAML and OIDC with Okta, Azure AD and Google Workspace, multi-user organization and seat management, API keys stored only as SHA-256 hashes, caller-based policy enforcement on every request, TLS 1.3 and encryption at rest with per-organization schemas, and audit log export to a SIEM, several of them on the Enterprise tier.

The compliance claim is hedged: the Enterprise page's answer to whether Temporal Cortex is SOC 2 compliant is that it is designed for SOC 2 Type II compliance, although a card on the same page reads SOC 2 Type II Compliance, and a HIPAA BAA is a contract offer, not an attestation. No SOC 2 report is published.

Sourcetemporal-cortex.com/enterpriseread 2026-09-22

On the managed platform, what agents do is recorded: an activity feed of booking and calendar activity per tenant, a dashboard and daily usage metrics counting bookings and API calls, and, on Enterprise, audit log export with SIEM integration and custom retention policies. The activity feed covers bookings rather than every read the agent made.

Sourcetemporal-cortex.com/docs/rest-apiread 2026-09-22

Settings, booking rules, caller profiles with trust levels and policies are stored, but the platform applies them as rules to each request rather than as context an agent reads to decide, and its distributed locks last only for a booking. No memory layer that keeps an agent's context across runs is documented.

Sourcetemporal-cortex.com/docs/rest-apiread 2026-09-22

Three deployment modes are documented: a local MCP server started with npx that keeps credentials on the user's machine, the managed platform, and a full self-hosted platform (API, MCP server, database and portal) through Docker Compose or Terraform modules on AWS, GCP or Azure; Enterprise customers can choose US or EU data residency.

Sourcetemporal-cortex.com/enterpriseread 2026-09-22

The vendor's own Agent Skills (router, datetime and scheduling) load a ready-made scheduling workflow into Claude, Cursor and other agent platforms, and Enterprise admins can provision skills and policies to users. These are working assets for one workflow; there is no browsable catalog of ready-made workflows, templates or role specific agents, and the use case pages (recruiting, sales, healthcare and others) are guides, not assets a buyer adopts.

Sourcetemporal-cortex.com/llms-full.txtread 2026-09-22

Scheduling requests reach a Temporal Cortex user through several channels: a Temporal Link that serves a booking page to people and JSON to agents, A2A Agent Cards for agent-to-agent negotiation, MCP tools and the REST API, with the owner's policies deciding each request. Those requests are handled by the platform's rules, and nothing documented starts the customer's own agent on an event or a schedule; the webhooks in the API reference are inbound sign-up and billing events for Temporal Cortex's own operations.

Sourcetemporal-cortex.com/open-schedulingread 2026-09-22

The tools work with whatever model and agent the customer runs: Claude Desktop, Cursor, OpenClaw, Manus or any MCP client, and LangGraph, CrewAI and the OpenAI Agents SDK in the developer guide, each with the customer's own model. Temporal Cortex names no model of its own and computes time deterministically rather than through a model, so model choice is the customer's throughout.

Sourcetemporal-cortex.com/developersread 2026-09-22

Outside callers use a documented REST API with Bearer API keys (calendars, availability, bookings, events, settings, callers, policies, keys and usage), an MCP server over stdio and streamable HTTP, A2A for agent-to-agent negotiation, and packages on npm, crates.io and PyPI, with TypeScript and Python SDKs named in the vendor's llms.txt.

Sourcetemporal-cortex.com/docs/rest-apiread 2026-09-22

Temporal Cortex publishes its own comparison of frontier models against its Truth Engine on recurrence rules and states its own test counts; both describe the vendor's product, not a way for the customer to test its agent. No fixtures, datasets, scoring or quality gate for the customer's agent workflows is documented.

Sourcetemporal-cortex.com/llms-full.txtread 2026-09-22

Temporal Cortex's browser channel is a booking page where people pick a slot on a Temporal Link, and its tools reach calendars through provider APIs. No browser, desktop or computer control by an agent is documented.

Sourcetemporal-cortex.com/open-schedulingread 2026-09-22