# Custom Instructions > Choose how Ditto responds, for your account or one Ditto agent. Find the settings, manage personality and writing preferences, and keep selected topics in sync with memory. --- # Custom Instructions Custom instructions let you choose **how Ditto responds**: your preferred tone, answer length, language, priorities, and context you want it to consider. Set your everyday preferences once for your account, then give a particular Ditto agent its own instructions. **Looking for the controls?** Open [Settings → Interface](https://app.heyditto.ai/settings/preferences). **Custom instructions** is at the top, above language and theme settings. In **Apply to**, choose **Your account** or a Ditto agent. The controls currently live here, including instructions for individual agents. ## Think of instructions as a set of cards Each named entry is one piece of your setup. You can edit it or turn it on and off independently. If you think of these as “mods,” a personality card is one mod and a writing-style card is another. In the app, the feature is called **Custom instructions**, and **Add instruction** creates another entry. | Instruction card | Example purpose | Where it belongs | | ------------------- | -------------------------------------------------------------------- | ---------------------------------- | | Personality profile | Use your existing personality assessment as background for responses | Your account | | Writing style | “Start with a short answer. Add detail when I ask.” | Your account | | Language tutor | “Correct my Spanish gently and explain one mistake at a time.” | A tutoring Ditto | | Project context | Consider supported memories about a particular project | Your account or the relevant Ditto | These cards guide responses. They do not install software, connect an app, or give an agent new tools. For those capabilities, see [Connected Apps](/docs/connected-apps) and [MCP Connections](/docs/mcp-connections). ## Your account and individual Dittos **Your account** is your personal default: the preferences you want across supported personal Ditto conversations. **An individual Ditto** adds preferences for that agent's role. Manage both from **Apply to**. **Main Agent** means only your main conversational Ditto; **Your account** is the broader default. The selector currently lists Dittos using your main memory; agents with a dedicated memory graph are not listed. For example, put “Use plain language and metric measurements” on your account. Give your cooking Ditto “Include ingredient amounts and a numbered method.” The cooking Ditto can use both. Your edits take priority over learned preferences. For the same instruction, an explicit edit takes priority over an inferred value; between equally explicit versions, the agent-specific version takes priority over the account version. Turning off the agent version of the same instruction suppresses that account instruction for the agent. Giving two new cards the same display name does not link them into an override. Avoid contradictory cards: an unrelated new card is an addition, not a replacement. Check **Effective instructions (saved)** after saving to see what is included for the selected scope. Your current request still takes priority, and preferences cannot waive approvals or security rules. Personal instruction layers are not applied to shared company conversations. ## Add your first instruction 1. Open [Settings → Interface](https://app.heyditto.ai/settings/preferences). 2. Under **Custom instructions**, choose **Your account** in **Apply to**. 3. Leave **Use custom instructions** on. 4. Select **Add instruction** and give it an **Extension name**, such as “Writing style.” “Extension” here means an instruction card. 5. In the text box, write the behavior you want. For example: “Start with the answer. Use short paragraphs. Explain unfamiliar terms.” 6. Leave **Keep a topic in sync** empty for a manual instruction. 7. Select **Save instructions**. To add instructions for an agent, select that Ditto in **Apply to** and follow the same steps. Save your changes before switching: **Apply to** is disabled while you have unsaved edits. This changes which layer you edit; it does not create an agent or change its tool access. Use concrete preferences. “Include a short recommendation before the details” is easier to apply than “Be better at explaining things.” ## What is the Personality profile card? An existing [personality assessment](/docs/personality) appears as a default instruction card. It is one part of your setup, not the container for every preference. Keep it and add separate writing, interaction, or project cards. The current editor may show the assessment as structured text with labels such as Big Five, DISC, and MBTI. You do not need to understand or rewrite that formatting to add another instruction. Turn the card off if you do not want it used. Editing the card makes it your explicit instruction instead of a profile that follows later assessment updates. ## Let Instruction keeper maintain preferences On **Hero and Legend**, Instruction keeper can periodically search your memories for supported preferences. Automatic maintenance is enabled by default when the feature is set up, with a **Weekly** schedule. Choose **Daily**, **Weekly**, or **Monthly**, or turn off **Enable Instruction keeper** to stop scheduled updates while keeping saved instructions in use. Learning depends on evidence, not how many memories you have. A direct statement such as “I prefer brief answers with concrete examples” can be useful. Hundreds of unrelated memories do not establish a writing preference. Evidence can be in any language. Keeper maintains dated source statements alongside your instructions. It does not rewrite your explicit instruction text or edit your original memories. Unsupported or conflicting evidence can leave a topic without a learned result. Source statements older than 30 days stop being used, so a previously learned result may disappear. ### Keep a topic in sync 1. Add or edit a card and write the instruction you want Ditto to follow. 2. In **Keep a topic in sync**, describe what to look for in memory. 3. Choose **Maintain as**: **Response preferences** for how to answer, or **Background context** for facts to consider. 4. Save, then let the next scheduled check run. For a “Family projects” card, you might write: - **Instruction:** “Consider my family's current projects when relevant. Ask when details are missing.” - **Topic:** “My close family and their current projects.” - **Maintain as:** Background context. That topic requests evidence; it is not evidence itself. Keeper must find a supported source before adding a family or project fact. It will not invent a relationship from the card's name. You can maintain up to three custom topics per layer. Current maintenance keeps a small set of short, complete source statements; long, quoted, or ambiguous memories may produce no result. Treat dated context as background, not a guarantee that someone's plans are still current. ## Understand the controls and status | Control or status | What it means | | ------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Use custom instructions | Turning this off for an agent disables both its own and inherited account instructions in that agent. Turning it off for your account disables personal instruction layers across your agents. | | Enable Instruction keeper | Whether automatic updates run. Turning it off does not erase saved instructions. | | Checkbox beside a card | Whether that individual instruction is used. | | Update frequency | How often Keeper checks eligible memories. Changing it schedules the next check using the new frequency. | | Pending / Next check | A check is scheduled; it has not yet produced a result. The time can pass before a worker picks it up. | | Completed | A check found supported evidence. Inspect the card's sources to understand the result. | | Abstained / Conflicting | There was not enough suitable evidence, or the evidence disagreed. An empty result is intentional. | | Error | The update failed. The previous saved result is retained and a retry is scheduled. | | Effective instructions (saved) | A technical view of saved instructions and evidence sent to Ditto. Unsaved edits are not reflected here. | The effective view currently includes structured data and a fixed safety explanation. You do not need to edit or copy it. Change the cards above and save them. The **Live preview** beside Interface settings previews appearance, such as fonts and spacing; it does not demonstrate how custom instructions change an answer. If you cannot find an expected result, check **Apply to**, both enable switches, the card's checkbox, and whether you saved. For automatic updates, also check your plan, the next check time, and the sources shown on the card. Manage the settings through this editor; chatting with Instruction keeper does not save settings changes. ## Related guides - [Settings Reference](/docs/settings) - [Personality Insights](/docs/personality) - [Memory Network](/docs/memory-network)