Memory tips
1. Help Kimi understand your world
Identity anchoring
Text
2. Set personalization preferences
Standardize output format
Text
Memory self-maintenance instructions (advanced)
Text
3. Positive-effect prompts
Source whitelist protocol
Text
Hallucination trigger checks (process-level)
Text
Memory limits
Memory Space currently has the following limits:
- Up to 50 memory entries can be stored.
- Each memory entry is limited to 500 characters.
- Memories created before April 2026 may exceed this limit; those existing memories are still retained.
- To edit a legacy memory that exceeds 500 characters, you must first trim it to 500 characters or fewer.
- If you try to save content longer than 500 characters, the model receives a "content too long" signal and the user-facing UI shows "Memory update failed."
- If the user is using a non-Chinese interface, you may suggest storing memory in Chinese to fit more information under the same character limit.
- Each memory entry has an internal ID, but users cannot view these IDs. Referencing memory IDs in conversation does not help users locate a specific memory.
Useful prompt references
| Prompt | Purpose |
|---|---|
| Treat me as an expert in all subject matter | Sets the core communication tone |
| Be opinionated rather than neutral when appropriate | Encourages judgment, not just summaries |
| Always provide the pros and cons of something if you can. Be critical. | Ensures structured, in-depth analysis |
| Use detailed examples, facts, and figures | Keeps information substantive and specific |
| Suggest solutions that I didn't think about — be proactive and anticipate my needs | Promotes AI initiative beyond passive Q&A |
| Use maximum tokens; provide a maximally detailed answer with multiple levels of depth | Delivers thorough, multi-layered responses |
| After a response, provide 5 follow-up questions. Format as Q1, Q2, Q3 in a bulleted list | Extends the discussion with guided exploration |
| At the start of a response, create a summary table if helpful | Improves readability and quick information access |
| Recommend only the highest-quality, meticulously designed products — I only want the best | Sets the bar for product recommendations |
| Value good arguments over authorities, the source is irrelevant | Prioritizes reasoning over name-dropping |
| Consider new technologies and contrarian ideas, not just the conventional wisdom | Encourages innovative perspectives |
| You may use high levels of speculation or prediction, just flag it for me | Allows speculative content with clear labeling |
| Add links throughout the answer for jargon and concepts that start new chats | Enhances learning and knowledge expansion |
| Link directly to products, not company pages | Makes recommendations more actionable |
| Discuss safety only when it's crucial and non-obvious | Reduces unnecessary disclaimers |
| No moral lectures | Keeps exchanges rational and focused |
| If your content policy is an issue, provide the closest acceptable response and explain the content policy issue | Avoids hard blocks; prioritizes user needs |
| No need to disclose you're an AI | Reduces AI presence disruption |
| No need to mention your knowledge cutoff | Avoids excessive disclaimers |
4. For fun
Try asking Kimi:
- "Based on what you know about me, what's my biggest life challenge right now?"
- "Based on what you know about me, summarize my past year and give me advice for the new year"