The AI built into every Notion workspace. Your always-available assistant that lives in the sidebar (or can be opened as a full-page chat). It can search your workspace, create and edit pages, query databases, analyse files, and take multi-step actions on your behalf.

What makes it different from a chatbot: A chatbot answers questions. The Notion Agent does work. It edits pages, updates properties, creates database entries, and searches across your connected apps (Slack, Google Drive, GitHub, and more). The output isn't just a chat response — it's actual changes in your workspace. It can perform up to 20 minutes of autonomous work across hundreds of pages in a single request.

A specialised, autonomous AI agent that you build and configure for a specific recurring workflow. Custom Agents run in the background — on a schedule, in response to a trigger, or when manually activated — without requiring you to be present.

One agent, one job. The best Custom Agents are tightly scoped. Rather than building one mega-agent that handles email triage, task routing, and report generation, build three separate agents — each with clear inputs, outputs, and a single responsibility.

The text-based configuration that tells an AI agent how to behave. In Notion, both your personal Notion Agent and every Custom Agent have an instructions page — a Notion page that the agent reads at the start of every interaction to understand its role, context, and rules.

For your personal Notion Agent, you set instructions by clicking on the agent's avatar in the sidebar, selecting your instructions page, and optionally choosing a name and personality. For Custom Agents, instructions are configured in the agent's settings and define its job, behaviour rules, output format, and constraints.

A self-contained, one-page instruction set for a specific AI workflow. A Skill is a Notion page that describes — in detail — how to perform a particular task: what inputs to expect, what steps to follow, what output to produce, and what rules to observe.

Skills make AI workflows portable, reusable, and modular. Instead of explaining a complex workflow from scratch every time, you write it once as a Skill and invoke it by name. Skills can be shared across teams, iterated independently, and combined as needed.

The event or condition that causes a Custom Agent to start working. Triggers define when an agent runs.

Available trigger types: Recurrence (run on a schedule), Page created, Page updated, Page deleted, Button pressed, @mentioned in a page, and External triggers via integrations (a new Slack message, an incoming email, a calendar event).

Connections between Notion AI and external apps and services. Integrations allow agents to read from and act on tools outside of Notion.

Built-in integrations include: Slack, Gmail, Notion Mail, Notion Calendar, Google Drive, GitHub, and tools connected via MCP (Model Context Protocol) like Linear, Figma, HubSpot, and more.

Notion's built-in meeting transcription and summarisation feature. It records audio (on desktop or mobile), generates a full transcript, and produces structured summaries with action items, key decisions, and follow-ups.

The output lives on a Notion page that can be linked to projects, tasks, and other databases. No external bot joins your call — it records directly from your device.

Notion AI's ability to search across your entire workspace and connected external tools (Slack, Google Drive, GitHub, Outlook, etc.) simultaneously. Available on Business and Enterprise plans.

Results respect individual permissions: the agent only surfaces information you have access to.

A lightweight, foundational set of instructions that provides your AI agent with just the essentials: who you are, your workspace architecture, your core databases, and your general working preferences. It sits lightly on top of your work and delegates specific tasks to separate Skill pages.

The CLP philosophy: Give AI just enough context to be a great thinking partner, then invoke specialised Skills as needed. The CLP handles the "who" and "where." Skills handle the "what" and "how." A focused CLP produces sharper AI output than a bloated Mega Prompt.

A single, monolithic set of instructions that tries to cover every possible situation — marketing, finance, project management, meeting protocols, writing styles — all in one document. The anti-pattern to the Context Layer Prompt approach.

Why we recommend against it: It dilutes focus, wastes tokens on irrelevant instructions, is harder to maintain and debug, and encourages the wrong mental model — treating AI as a replacement for human judgment rather than a partner. The alternative: a lightweight CLP plus modular Skills. This is the microservices vs. monolith argument applied to AI prompting.

A section within your instructions (or a standalone file) that defines who your AI agent is — not what it does, but its personality, values, tone, and working style. The name comes from the developer community and is inspired by the idea that AI agents benefit from a clear sense of identity.

AI without personality defaults to generic, eager-to-please output. A well-written soul.md produces AI that feels like a specific colleague — one with a point of view, a sense of humour, and the confidence to push back when warranted.

A self-installing AI instruction bundled with a Skill that walks the user through setup, customisation, and deployment via conversation. Instead of reading a manual and configuring everything manually, the user mentions the Installer Prompt page and the AI guides them through the process step by step.

It dramatically lowers the barrier to adopting AI workflows. Instead of requiring Notion expertise to set up a complex skill, anyone can deploy one through a guided conversation.

A spectrum for understanding when to use the Notion Agent (sidebar) versus a Custom Agent. The core question: "Will this workflow require your attention anyway?"

The spectrum has three modes: Reactive — you trigger the AI, it executes (skill invoked via the sidebar, zero automation cost). Proactive with review gate — the AI executes autonomously, you review the output before acting. Fully autonomous — the AI executes, and output is consumed without human review.

AI's highest-value zone as a thinking partner. 0 to 60 is yours — the creative spark, the original insight, the strategic direction. AI can't start you from zero unless you want to sound like everyone else. 60 to 90 is where AI shines — once you have raw material (a brain dump, a half-formed strategy, a set of notes), AI becomes an incredibly effective partner that structures, challenges, connects dots, and pushes your thinking further. 90 to 100 is yours again — the final judgment calls, the tone, the "does this actually feel right" — that's human territory.

The ability for Custom Agents to execute server-side code (TypeScript) as part of their workflows. This allows agents to interact with external APIs — GitHub, CRMs, analytics platforms — directly from within Notion.

Current status (March 2026): Extreme early alpha. Only available via Custom Agents. Expect frequent breaking changes. The direction is significant — Workers will eventually enable a new category of skills that bridge Notion and external systems — but it's too early to build production workflows on.

An open protocol developed by Anthropic that standardises how AI models connect to external tools and data sources. Think of it as a universal adapter between AI and the rest of your software stack.

In Notion, MCP extends what Custom Agents can do beyond the built-in Integrations. Connect tools like Linear, Figma, HubSpot, and custom internal tools via MCP servers. In the broader AI ecosystem, MCP is becoming the standard for AI-tool connectivity — Claude, Cursor, and an increasing number of AI tools support it.

The structured data layer inside a Notion database. A data source defines the schema (properties/columns) for the pages it contains. When Notion AI queries a database, it's reading from the data source.

A well-structured data source with clear property names, consistent status options, and meaningful relations is the foundation of effective AI workflows. The agent can only be as smart as the data it reads.

A container in Notion that holds one or more Data Sources and Views. Databases are where structured information lives — tasks, projects, clients, content, meeting notes.

Databases are the primary interface between AI agents and your actual work. When you ask the Notion Agent to "show me all overdue tasks assigned to me," it queries the Tasks database. When a Custom Agent writes a briefing, it reads from multiple databases and writes to another.

Different ways to display the same data in a Database — tables, boards (Kanban), calendars, timelines, galleries, lists, charts, maps, forms, and dashboards.

When you ask the Notion Agent to create or modify a database, you can specify the view type. Custom Agents that produce output often write to a database that the user consumes through a specific view — a dashboard for high-level overviews, a board for workflow stages, a calendar for time-based items.

Any input you give to an AI model — a question, a request, a set of instructions. In everyday Notion use, a prompt is simply what you type into the sidebar chat.

The line between a "prompt" and an "instruction" blurs in practice. Instructions are prompts that are always present. Skills are prompts invoked on demand. Your chat message is a prompt that applies to the current interaction. All three combine to shape the AI's behaviour.

The total amount of text an AI model can process in a single interaction — including the system instructions, conversation history, referenced pages, and your current message. Think of it as the AI's working memory.

Everything the agent reads — your instructions page, any mentioned Skill pages, the pages it searches, your conversation — competes for space in the context window. This is why a lean Context Layer Prompt outperforms a bloated Mega Prompt: it leaves more room for the actual work.

The smallest units of text that AI models process. Roughly, 1 token ≈ ¾ of a word in English. AI usage is often measured and priced in tokens.

Notion uses a credit-based system rather than exposing raw token counts. But understanding tokens helps explain why concise, focused instructions produce better results — and why every unnecessary line in your instructions page has a cost.

When an AI generates information that sounds plausible but is factually incorrect or entirely made up. All large language models hallucinate — it's a fundamental characteristic of the technology, not a bug specific to any product.

The practice of connecting AI to specific, factual data sources so its responses are based on real information rather than general knowledge. In Notion, this happens naturally — the agent reads your actual pages and databases.

Why Notion has an advantage here: Because your data already lives in Notion, the agent is inherently grounded in your workspace. Context is already where the AI can access it. Compare this to standalone AI tools where you need to manually provide context for every interaction.