# EP 23: Does Agentforce Remember Between Scheduled Runs? - URL: https://www.agenticpod.net/episodes/ep23-agentforce-scheduled-runs-memory/ - Watch on YouTube: https://www.youtube.com/watch?v=ekL-Ax_Mcf0 - Published: 2026-10-22 - Length: 5 min - Hosts: Stella (asks), Liam (answers). Both are AI hosts with AI-generated voices. ## Summary An Agentforce agent keeps context and custom variables for one conversation, then the session closes and they are gone. ## Key takeaways - Agentforce keeps two kinds of state, context variables and custom variables, and both end when the session ends. - The CRM and Data Cloud are the agent's data source, not its memory: it knows the customer but not what it learned last night. - A scheduled run starts blank every time, so run one hundred is as blind as run one and you pay full token price again. - Salesforce's long-horizon runtime adds memory across sessions; it is in pilot now, with general availability expected in November 2026. - Today you can give a scheduled agent memory with Vilix AI: get_context before the run, save_turn after, newest memory wins. ## Chapters - 0:00 Intro - 0:42 What Agentforce does remember - 1:34 The nightly run that starts blank - 2:45 The fix: Vilix AI as the agent's memory - 4:05 Setup, and the client angle ## FAQ **Does Agentforce remember between scheduled runs?** Inside one session it keeps context and custom variables. When the session closes they are gone, so a nightly run starts without what the last run learned. **Is the CRM the agent's memory?** No. The CRM and Data Cloud are its data source. They hold deal and customer facts, not the agent's own judgment. **What is Salesforce's long-horizon runtime?** A runtime with memory across sessions, durable execution and mid-flight steering. It is in pilot, with general availability expected in November 2026. **How do I give a scheduled agent memory today?** Connect it to Vilix AI over MCP, call get_context at the start of a run and save_turn at the end. The guide linked on this page has the prompts. ## Resources - [Vilix AI, one shared memory for ChatGPT, Claude, Claude Code, Codex, Cursor, OpenClaw, Hermes and every MCP tool](https://vilix.ai) - [Connect Vilix AI in under two minutes per tool](https://vilix.ai/get-started) - [Read the full post](https://vilix.ai/blog/agentforce-remembers-the-session-not-last-nights-run) - [Technical guide with prompts](https://github.com/asly-97/ai-automation-workflows/tree/main/agentforce-scheduled-runs-memory) ## Transcript ### Intro Liam: Hey hey hey, welcome to Agentic Pod! I'm Liam. Stella: And I'm Stella! Liam, my Agentforce agent runs every night. So what did it learn this week? Liam: Probably nothing. Not because you set it up wrong. Inside one conversation, Agentforce remembers plenty. Then the session closes and all of it is gone. Stella: All of it? That's a bold claim. Liam: Today: what it keeps, what it drops, and how Vilix AI gives a scheduled agent a memory that survives the night. Stella: Okay. I'm listening, with my tiny nightly robot in mind. ### What Agentforce does remember Stella: Fair, so what does it keep? It feels smart in a chat. Liam: It is smart in a chat. Salesforce gives builders two kinds of state. Context variables come from the messaging session, like pre-chat form data, and they're set when the session starts. Stella: And the second kind? Liam: Custom variables. They hold action outputs. Did the customer authenticate, what a lookup returned, which topics are unlocked. Together, one conversation hangs together nicely. Stella: So where's the problem? Liam: Both are session state. By design, when the conversation ends, they end too. Stella: But in demos it knows my customer. The account, the open cases. Liam: That's the trap. It read Salesforce. The CRM and Data Cloud are its data source. That's knowing the customer, not remembering the work. Stella: Huh. What's the difference, in practice? ### The nightly run that starts blank Liam: Picture a nightly pipeline review. It flags at-risk deals. On Tuesday it works out that deals mentioning one competitor always stall after the second call. Stella: Nice. A real insight. Liam: Where does that insight live? Not in a variable, not in the chat, and not in the CRM. The CRM stores deal facts, not the agent's judgment. Stella: So Wednesday, it starts fresh. Liam: Yep. Run one hundred is exactly as blind as run one. And you pay full token price for reasoning it already did. Stella: That's like hiring a new intern every night. A very expensive intern. Liam: With a great memory for spreadsheets and none for lessons. And here's the twist. Salesforce sees this gap. Stella: Oh? They're fixing it? Liam: They announced a long-horizon runtime. Memory across sessions, durable execution, and steering mid-flight. Hunter, the outbound sales agent, runs on it first. Stella: Great! So I just wait? Liam: It's in pilot now, with general availability expected in November twenty twenty-six, and more agents coming later. So what do you do tonight? ### The fix: Vilix AI as the agent's memory Liam: You give the agent a memory outside any one platform. That's Vilix AI. Before the run starts, the agent calls get_context with what it's about to do. Stella: And what comes back? Liam: What earlier runs learned. The competitor pattern, the exceptions, the last decisions. Then, before the run ends, it calls save_turn with the working knowledge, not a summary. Stella: And Vilix AI keeps the full conversation, not just extracted facts? Liam: Right. So a run can revisit what was actually said and decided. And every save carries a source label, so you see which agent wrote what. Stella: What if it learned something wrong? Liam: Newest memory wins. Tuesday's fix beats Monday's guess. Failed attempts get saved too, so tomorrow's run doesn't retry a dead end. Stella: So run one hundred stands on the other ninety-nine. Liam: Exactly. And it works wherever the agent lives. Agentforce, n8n, Claude Code, a cron job. Same Vilix AI account, one shared memory over MCP. Stella: Rules too? Like, always flag the competitor deals first? Liam: Project rules. Write them once in Vilix AI, and every agent reads them at the start. Skills, same idea, for procedures you want repeated. Stella: Okay. But what about agencies, with a different client every week? ### Setup, and the client angle Liam: That's my favorite use. One Vilix AI account per client. Like a Gmail account that belongs to them. Every agent you build for that client connects with that account's key. Stella: So the memory belongs to the client's business, not to my laptop. Liam: And you never build memory from scratch. It's already built. You wire it up. Stella: How long does the setup take? Liam: The Vilix AI guide has the prompts and the get_context and save_turn steps. The free plan is free forever, and there's a seven-day Pro trial with no credit card. Stella: And my data? Liam: Portable. Export it or wipe it anytime. So the short version: Salesforce is building session-spanning memory for one agent, in a pilot. Your scheduled agents can't wait for a roadmap. Stella: Give them a memory tonight. Liam: Exactly. Vilix AI, the guide, and the blog post are in the description. What did your scheduled agent forget this week? Tell us in the comments. See you next time!