monday.com

Agents and the assistant act through the customer's connected tools, with Slack connected over OAuth, Gmail and Google Calendar named, and more than two hundred native integrations across Microsoft Teams, Google Workspace, Jira, GitHub, Salesforce, HubSpot and Zapier with two-way sync.

SourceActions reach the work itself as well: creating content, updating tasks and timelines, notifying teammates and triggering workflows from a prompt, and the platform's automation and integration engine carries the calls, monday.com monday sidekick, integrations and AI agents pagesread 2026-09-17

Agents are built in the agent builder in a few steps, given the knowledge and tools they need, and then act across the work rather than firing a single step: the named agents are described detecting a condition across projects and then taking several actions in response, reassigning owners, updating timelines and alerting stakeholders, and monday states agents keep work in sync across tools by pulling context and taking the actions that would otherwise be handoffs.

SourceDeterministic sequencing sits underneath in the platform's automations and AI workflows, and the assistant coordinates execution across the platform, monday.com AI agents, agent builder and monday sidekick pagesread 2026-09-17

The agent builder has knowledge as one of its steps, and monday states that agents use the docs, PDFs and boards the maker defines as context so that every action is grounded in the customer's own work and guidelines, with the assistant reading across boards, docs and connected tools.

SourceWhat holds it short of Full: no index, refresh behavior, per-source permissioning or citation surface is documented, so a buyer cannot see which documents influenced an output, monday.com AI agents and monday sidekick pagesread 2026-09-17

Guardrails on the AI Trust Center are platform safety controls: an AI gateway with rate limits and token limits, moderation and security, and named model providers under zero-retention terms. Admin settings and granular permissions define who can use AI across the account and by user role, agents are described as offering transparency into every action with guardrails and permissions that can be set, and the assistant states that it explains each action before taking it and that the user reviews and approves.

SourceWhat holds it short of Full: no per-action approval step or pre-execution policy check is documented at the agent layer, monday.com AI Trust Center, AI agents and monday sidekick pagesread 2026-09-17

monday publishes a Trust Center that carries the attestations as downloadable reports rather than on request, including a SOC 1 Type II report, a SOC 2 Type II report, a SOC 3 report, a HIPAA business associate agreement, ISO certification pages, and statements for GDPR and UK GDPR and the Australian Privacy Act. The customer-facing controls sit beside them: role-based access with granular permissions, single sign-on, two-factor authentication, admin audit logs exportable by CSV or API, IP allowlisting, and account-level control over who may use AI at all, with a separate acceptance step before AI is enabled.

Sourcemonday.com/trustcenterread 2026-09-17

Platform auditability is real: admin audit logs record account activity and export by CSV or programmatically through the API for archival in the customer's own systems, and there is a public status page with continuous monitoring.

SourceWhat holds it short of Full: no agent reasoning trace, per-agent decision log or tool-call record is documented, so what is auditable is what changed rather than why the agent chose to change it, monday.com Trust Center and AI agents pagesread 2026-09-17

What persists is the platform's own data model: agents read and write boards, items and docs on every run, so state lives in the work records rather than in the agent. Nothing documents a memory store, retention of context between runs, or any surface for a user or administrator to view, edit or delete what an agent has retained, and the assistant's privacy statements cover chats staying private rather than persisting as recallable memory.

Sourcemonday describes agents as learning from patterns over time, which is absorbed learning rather than a memory a user can inspect, monday.com AI agents, monday sidekick and AI Trust Center pagesread 2026-09-17

Hosting is customer-selectable across named regions, with data centers in the United States, the European Union and Australia and a disaster-recovery site in a separate region, and the residency choice propagates: monday states that AI follows the same data residency policies as the account so AI processing and storage stay inside the designated region, and apps hosted on monday code inherit residency matching the account's region.

SourceThe limit: it is explicitly cloud only, with no on-premises or self-hosted option, monday.com AI Trust Center, Trust Center and monday code developer documentationread 2026-09-17

Agents come ready-made or custom. monday names individual prebuilt agents with the work each does, among them a Risk Analyzer that detects schedule, dependency and workload risks across projects and then reassigns owners, updates timelines and alerts stakeholders, and an RSVP Manager that tracks invites and responses and sends reminders, alongside expert agents grouped for sales, service and PMO, department agents for marketing, IT, HR and operations, and an AI template center.

SourceThe agent builder creates custom agents from a prompt in a few steps, so a buyer both selects finished units and composes new ones, monday.com AI agents page, agent builder and AI template centerread 2026-09-17

Routines are scheduled AI actions that run automatically on a set cadence, configured from the assistant's sidebar, so work reaches the agent without a person asking.

Sourcemonday.com/w/sidekickread 2026-09-17

monday discloses that its AI runs on managed model APIs from Microsoft Azure and AWS Bedrock under zero-retention terms, with the providers not storing or training on inputs or outputs, and states that customer input and output are not used to train machine-learning models. The assistant is described as bringing together several models. That routing is the vendor's, not the customer's; no model selection, per-agent assignment or bring-your-own-model surface appears, and the AI Gateway is documented under guardrails as rate limits, token limits and moderation rather than as a model picker.

Sourcemonday.com/w/ai-trust-centerread 2026-09-17

The platform is callable and extensible from outside: a documented GraphQL API with its own reference site, monday code as a hosted runtime for apps with data residency matching the account region, a Developer Center and marketplace where publishers attest to data segregation, least privilege and penetration testing, and an MCP server installed from the marketplace and scoped to the installing user's permissions and chosen workspaces.

Sourcemonday also publishes machine-readable skill files and a programmatic signup flow so an external agent can obtain its own account and API token, monday.com developer documentation, monday MCP page and published agent skill files; developer.monday.com/api-reference/docsread 2026-09-17

Nothing documents a test environment, a simulation harness, a scored evaluation or an agent debugging surface. The agent builder's flow is authoring: a few steps to define the agent, its knowledge and its tools, with the prompt reviewable before launch, and the guardrails monday publishes are rate limits, token limits and moderation rather than measurement of agent behavior.

SourceNo before-deployment test run and no after-deployment quality score is described, monday.com AI agents, agent builder and AI Trust Center pagesread 2026-09-17

Agents act through the platform's own records, its automations, its integrations and the GraphQL API, and no browser, desktop or computer control is documented. The one place a browser is mentioned is a signup page for external agents rather than an agent operating an interface, so the question of what happens when a screen changes does not arise for this product.

Sourcemonday.com/w/agentsread 2026-09-17