> ## Documentation Index > Fetch the complete documentation index at: https://docs.every.ai/llms.txt > Use this file to discover all available pages before exploring further. # Memory & Skills > How Every Agent remembers facts across chats and applies the reusable skills you create to shape how it works ## Overview Every Agent doesn't start from scratch in every conversation. Two systems let it carry knowledge forward and apply your preferred way of working automatically: * **Agent Memory** — the agent keeps a private, evolving notepad of facts and observations about you, available in every chat. * **Skills** — reusable guidance documents you create that tell the agent how to handle specific kinds of work. Create, enable, and edit your agent's skills Context, voice, sources, and preferences ## Agent Memory As you work with Every Agent, it builds up a running understanding of you and your business — open todos, project details, things you've mentioned, observations worth keeping. The agent writes these down to its own **memory** so they persist from one chat to the next. **How it works:** * The agent decides what's worth remembering and saves it as it goes — you don't have to manage memory yourself * That memory is loaded into every conversation, so the agent has its notes in view from the first message * Memory is **private to you** — in a shared workspace, teammates do not see each other's agent memory **What lands in memory:** * Running lists and open todos ("still need to follow up with ACME on the renewal") * Observations about how you work or what you prefer * Project and client details the agent picked up in earlier chats You don't need to do anything to turn memory on. If you want the agent to remember something specific, just tell it — "remember that the Q3 project kicked off in June" — and it will note it. ### Memory vs. Preferences vs. Context Every Agent has a few different ways of knowing things, and they serve different purposes: | Surface | What it holds | How it's set | | ----------------------- | --------------------------------------------------------- | ---------------------------------------------- | | **Agent Memory** | Freeform recall — running lists, observations, open todos | The agent writes it as you work, automatically | | **Learned Preferences** | Standing rules — "always do X," "never do Y" | The agent proposes a rule; you confirm it | | **Business Context** | The narrative of what your business is and does | You write it, or build it from your workspace | They reinforce each other: Memory tracks the loose ends, Preferences enforce hard rules, and [Business Context](/ai/context-and-voice) gives the big picture. Manage standing rules in [**Agent → Preferences**](https://every.ai/agent/preferences); manage Business Context in [**Agent → Context**](https://every.ai/agent/context). ## Skills **Skills** are short guidance documents that tell Every Agent how *you* want certain tasks done. Instead of repeating the same instructions every time, you capture them once as a skill, and the agent applies it automatically whenever it's relevant. A skill is plain language — there's no code involved. For example: * An **Invoice Formatting** skill that always adds your payment terms to the notes field and uses formal descriptions * A **Client Communication** skill that sets the tone and formality of outbound messages * A **Gmail** skill that defines how the agent should triage and reply to email Create and manage your organization's skills ### How Skills Get Applied Active skills are available to the agent in every chat. When your request matches a skill's purpose, the agent reads the skill and follows it — **you don't have to mention the skill by name**. ```text theme={"system"} You: "Create an invoice for ACME Corp for $2,000" → The agent applies your Invoice Formatting skill automatically: payment terms in the notes, formal line-item descriptions, your preferred format. ``` You can turn individual skills on or off, so you control exactly which guidance is active. You can also **@-tag** an active skill directly — in a chat message or in a [Scheduled Task's](/scheduled-tasks/scheduled-tasks) instructions — to make sure that skill's guidance applies to that request. ### Creating a Skill There are several ways to create a skill, from fastest to most hands-on: Click **Create** in the skills area and the agent interviews you — what kind of work is this for, what should it do differently, any examples — then writes the skill for you. This produces the highest-quality skills because the agent shapes your answers into clear guidance. Add a recommended starter skill (like Gmail or Calendar management) with one click, then edit it to match how you work. Use the editor to fill in a name, when it should apply, your guidelines, and good/bad examples — or write the markdown directly. Open a skill and choose **Tweak with AI** to refine it conversationally — add rules, adjust when it applies, or add examples. ### Writing a Good Skill The strongest skills are specific and give the agent concrete examples to follow: * **When to apply** — the situations that should trigger the skill (creating invoices, drafting emails) * **Guidelines** — clear, actionable rules ("always include payment terms in the notes field") * **Examples** — a good example and a bad example so the agent can tell the difference * **Exceptions** — when the skill should *not* apply Keep guidelines actionable, not aspirational. "Use Net 30 terms and reference the PO number" is more useful than "be professional." ### Skills vs. Preferences vs. Context Skills sit alongside the other ways you teach the agent: * Use a **Skill** for *how to do a kind of task* (how to format invoices, how to draft a proposal) * Use a [**Learned Preference**](/ai/every-agent) for a *single standing rule* ("ACME always gets Net 15") * Use [**Business Context**](/ai/context-and-voice) for *what your business is* (services, pricing logic, who you serve) ## A Note on Credits Memory and skills are part of how the agent works in chat — agent conversations consume credits. See [What is a credit?](/plans/what-is-a-credit) for how that works. ## Next Steps See how the agent uses memory and skills day to day Teach the agent what your business is and how it sounds Put skills and memory to work in recurring and one-time runs Give the agent more to reach with external connections **Need help?** Contact support at [hello@every.ai](mailto:hello@every.ai)