> ## Documentation Index > Fetch the complete documentation index at: https://docs.runspark.ai/llms.txt > Use this file to discover all available pages before exploring further. # Agent Memory > How your agent remembers things — across conversations, across days, and over time One of the first things people ask about their Spark agent is: *"Will it remember what I told it yesterday?"* **Yes.** Your agent has a memory system specifically designed to remember things — not just within a single conversation, but across days, weeks, and months. Let's break down how it works. *** ## The Three Layers of Memory Your agent's memory works in layers, similar to how human memory works: ### Layer 1: Conversation Context (Short-Term) This is what your agent remembers within your current conversation. If you say "I'm talking about the Johnson account" at the beginning of a chat, your agent remembers that context for the rest of the conversation. You don't need to repeat it. > **You:** I'm working on the Johnson proposal. > > **Agent:** Got it — pulling up the Johnson account details. > > **You:** What's their current contract value? > > **Agent:** The Johnson account is currently at \$8,500/month on the pro plan. Their renewal is coming up in June. > > **You:** Bump it to the enterprise tier and add the premium support add-on. > > **Agent:** Updated the Johnson proposal to the enterprise tier ($12,000/month) plus premium support ($2,500/month). New total: \$14,500/month. Want me to draft the proposal? Your agent kept track of "we're talking about the Johnson account" the entire time — you didn't need to repeat it in every message. **How long it lasts:** For the duration of your current conversation session. If you close the chat and come back later, the agent starts a fresh conversation (but reads its memory notes to recall recent context). ### Layer 2: Daily Notes (Medium-Term) Every day, your agent writes a journal entry — a log of what happened, what decisions were made, and what's important to remember. Think of it like a work diary. A daily note might look something like: ``` March 27, 2026 - Processed the Johnson renewal. Moved them to enterprise tier + premium support. $14,500/mo. Proposal drafted and in Work tab. - New lead: CoreTech Solutions. Enterprise inquiry for 200+ users. Flagged for review — above standard pricing tiers. - Correction: seasonal discount is June-August ONLY, not May-September. Updated my rules. - Owner prefers bullet-point summaries over paragraphs for morning reports. Noted for future formatting. ``` **How long it lasts:** Permanently. Your agent keeps one file per day, and it reads yesterday's and today's notes every time it starts a new session. This is how it picks up where it left off. ### Layer 3: Long-Term Memory (Permanent) Your agent maintains a curated memory file — a living document of the most important facts, preferences, and lessons it's learned over time. While daily notes are raw logs, long-term memory is the distilled, organized version. Long-term memory includes things like: * **Key client information:** "Riverside Group is our largest client. Renewal in June. Always get the premium package. Primary contact: David Chen." * **Your preferences:** "Prefers bullet points. Likes concise morning reports. Wants to review all deals over \$10K personally." * **Business rules learned from corrections:** "Always round up to the next package size. Seasonal discount is June-August only. Never offer more than 15% discount without approval." * **Important context:** "We're in the middle of a pricing restructure — hold off on quoting enterprise rates until the new pricing guide is finalized." **How long it lasts:** Permanently — and your agent actively maintains it. Periodically, it reviews recent daily notes and promotes the most important information to long-term memory. Old or outdated information gets cleaned up. *** ## How Memory Works Day-to-Day Here's what happens in a typical day: **Morning — Your agent wakes up:** 1. Reads its personality file (who it is) 2. Reads its rules and instructions (what to do) 3. Reads yesterday's daily notes (what happened recently) 4. Reads today's daily notes so far (anything from earlier today) 5. Reads its long-term memory (key facts and preferences) Now your agent has full context. It knows who you are, what you've been working on, what rules to follow, and what happened recently. **During the day — As you work together:** * Your agent references its memory when relevant ("You mentioned last week that the Henderson deal is on hold — is that still the case?") * It notes important new information as the day goes on **End of day — Your agent writes its notes:** * Logs key decisions, corrections, new information, and important interactions * These notes become part of tomorrow's context *** ## Telling Your Agent to Remember Things You can explicitly ask your agent to remember something. It will write it down in a way that persists: > **You:** Remember this: the Martinez family gets a 20% loyalty discount on all bookings. They've been with us for 5 years. > > **Agent:** Noted! I've saved that the Martinez family receives a 20% loyalty discount on all bookings (5-year client). I'll apply this automatically when processing their bookings. Other examples: > "Remember that Sarah prefers email over phone calls." > "Remember that we're not taking new clients until May." > "Remember that the office is closed on April 18th for a company retreat." > "Don't forget — the quarterly review meeting is always the last Friday of the quarter." Everything you tell it to remember gets written to a file that survives across sessions. Even if you don't talk to your agent for a week, it'll still remember. *** ## How Your Agent Curates Its Own Memory Your agent doesn't just dump everything into memory — it actively manages what it remembers, like a well-organized person cleaning out their notes: ### Promoting Important Details When something comes up repeatedly or proves important, your agent promotes it from daily notes to long-term memory. For example: * You've corrected the same pricing rule three times → it's now prominently noted in long-term memory * A client's name keeps coming up → key details about them are saved for quick reference * You've stated a preference multiple times → it's recorded as a core preference ### Cleaning Up Outdated Info Your agent also removes things that are no longer relevant: * "We're not taking new clients until May" → Once May arrives, this gets removed * "Reminder: follow up with David on March 20th" → After March 20th, this gets archived * Old client details that haven't been referenced in months → Moved to daily note archives ### The Learnings System When your agent makes a mistake and you correct it, it doesn't just fix the immediate issue — it logs a **learning**. Learnings are like lessons the agent teaches itself: > **Learning logged:** "When calculating group pricing, ALWAYS round up to the next group of 5. I made this mistake on March 15th and was corrected." These learnings accumulate over time, making your agent progressively more accurate. You can see what your agent has learned: > **You:** What have you learned from our conversations? > > **Agent:** Here are my key learnings: > > * Round up to next group of 5 for pricing (corrected March 15) > * Seasonal discount applies June-August only (corrected March 18) > * Always schedule a call before sending quotes (new process, March 20) > * Use casual-professional tone in client emails (preference, March 21) > * Martinez family gets 20% loyalty discount (noted March 22) *** ## Memory Across Channels Your agent's memory is shared everywhere. If you tell your agent something via Telegram on your phone, it remembers it when you chat in the Spark portal on your computer. This means: * Tell it to remember a client detail on Telegram → it references that detail in the portal * Correct a mistake in Slack → the correction applies everywhere * Upload a knowledge file through the portal → your agent can reference it from any channel **The conversation threads are separate, but the memory is one.** Think of it like a person who remembers everything they've learned, regardless of whether they learned it via email, phone, or in person. *** ## What If My Agent Forgets Something? It shouldn't — but here's what to do if it seems like it has: This can happen in very long conversations where a lot of information has been exchanged. Just remind it: "Earlier I said the budget was \$5,000 — remember?" Your agent will re-read the conversation and pick it up. Check if your agent wrote it in its daily notes. You can ask: "What did you write in yesterday's notes?" If the information is missing, tell it again and say "remember this for the future." Rules should be in the agent's instructions file. Ask: "What are your current rules for \[topic]?" If the rule is missing, set it again: "New rule: always do X when Y happens. Write this down." Client details should be in CRM contact records or memory. If it's missing, tell it again with emphasis: "Remember: the Henderson account is our largest client, \$50K annually, renewal in September. Save this." **When in doubt, just tell your agent again.** It costs nothing to repeat important information, and your agent will make sure it's saved this time. Over time, as your agent's memory gets richer, you'll need to repeat things less and less. *** ## Privacy and Your Agent's Memory Your agent's memory is private to your workspace: * **Only you (and your team members) can see it** — memory is not shared outside your workspace * **Your agent doesn't share information between workspaces** — each workspace is completely separate * **You can ask your agent what it remembers** — there's no hidden memory. Ask "what do you know about me?" or "what's in your memory?" and it'll tell you * **You can ask your agent to forget things** — "Delete the note about the Henderson account" or "Clear your memory about my personal preferences" Your agent is transparent about its memory. Ask it anytime, and it'll show you what it knows.