Run Long (@RunLong) header
Run Long (@RunLong) profile picture

AI explorer · Founder @gedolabs — a growth system that actually knows you: assistant within, digital self without. Training a model that's yours alone. 🇸🇬

Singapore
Joined March 2023
Personal AI is having its moment. But an agent that only remembers the last chat is still a stranger every morning. Gedo is built for your life — goals, decisions, people, and the story you keep living — not just the context of one conversation. A day with Gedo ↓
A lot of AI is racing to do everything on its own. We're trying a different path: an AI that works with you. Here's one day of it: a morning run, Mom's birthday dinner planned on a voice call, and a visitor from Tokyo. Gedo is in beta. Watch, and tell us if we're wrong.
3
1
7
528
Gedo Memory(gedo.ai) placed 3rd on the AML Cycle 2 Commercial Products leaderboard (Text Memory), and 5th among closed-source methods. I'm grateful to our team, and to the AML and CSIG organizers. A few notes on what this meant for us. Why we entered At Gedo we're building a personal AI that remembers you over months and years. The memory layer is the product: what it keeps, what it lets go, and what it never says out loud. For a long time we measured it only on our own tests. AML gave us something we couldn't build ourselves: an independent benchmark where the organizers fix the answering model and the judge, so the only thing that differs between systems is memory. What the run looked like We decided to enter in late October and had our system online a few days before the deadline. The full suite then ran for about 72 hours. It surfaced problems our own tests never had: shards running out of memory under sustained load, idle connections being dropped, a write path that needed to survive a restart. All of them were on our side, and fixing them made the system sturdier. What the scores told us We did best on explicit fact recall, multi-hop reasoning, and time and event order. Relative to the field, our weakest area was epistemic safety and privacy. That's the one I care about most. A personal agent that recalls everything but doesn't know what to keep to itself isn't one you'd trust with your life. Since the competition, that's where our work has gone: forgetting, and what the agent chooses to disclose. Agent memory is still early, and shared, independent benchmarks are how the field gets better. Thank you, @AgentMemoryL.
2
1
1
6
239
AI memory should do more than recall what you said. It should organize, correct, and help you act. That’s what we’re building with Gedo: source-linked memories, nightly review for conflicts and relevance, and connections to your goals and actions. For personal AI, there’s another question: can you inspect and correct what it remembers about you—and decide what other AIs can access? Models and tools change. Your memory should remain yours. gedo.ai/features/memory
Memory is what allows Claude to learn from experience. In a new Engineering Blog post, we share how “dreaming” helps Claude organize what it’s learned, correct its mistakes, and find what it needs more easily the next time. anthropic.com/research/compo…
6
1
4
155
Claude Ultra:5小时限制取消了,恭喜你现在可以一次性把整周额度在一个下午烧完(这是兑水了?)。然后心满意足地等下周二。效率是提高了,就是提高的方式有点刺激。
1
1
3
136
人生只有一次?No Gedo 世界让分身先替你走一遍:回放你记下的那些年,把接下来两年过一遍,每个路口等你选,回来时带着这周就能开始的准备。 目前灰度测试中,陆续分批开放,有兴趣的可以关注:Gedo.AI
Look back on your past. Explore your future. Gedo World is an early world model for your avatar. It replays the years you've recorded, simulates another way of living, and brings what it learns back as a plan. In limited testing, rolling out in batches. gedo.ai/world
2
3
140
Run Long retweeted
How it works: 1. Retrace: it replays the years you've recorded. 2. Simulation: it lives a path on a real-world map and stops at each crossroads for you. 3. Journal: what it learned, and where you aren't ready yet. Every screen is the real product. gedo.ai/help/gedo-world
1
1
102
When we started Gedo, I didn't care much about benchmark scores. What I cared about was whether it actually remembers your life: the small thing you mentioned once, how your plans changed, who matters to you. But you can't improve what you don't measure, so we put our memory through LoCoMo and LongMemEval. 96% on one, 87.2% on the other. The second number is the one I keep looking at. We have work to do. We also made a choice that won't look good on a speed chart. Gedo takes about 4.4 s to write each turn, because it keeps your words, turns them into facts that point back to where they came from, and checks them against what you said before. It runs after the reply, so you don't wait for it. I'd rather it be slower and right. A few things I keep going back and forth on. I'd like to hear where you land: 1. Context windows keep getting bigger. Does that make memory less necessary, or more? Pasting your whole history into every prompt isn't remembering. It's rereading. 2. These benchmarks test whether an AI can recall a fact. None of them test whether it noticed that you changed. Which matters more to you? 3. A friend who remembers every word you've ever said would be unsettling. One who remembers what matters is a friend. Should AI memory forget on purpose, and who decides what? 4. Who should your memory belong to: the model you happen to use this year, or you? Reaching 95% on both won't mean memory is solved. It means we've reached the starting line. When we get there, we'll open-source the architecture and the evaluation. If you're building in this space, or just have opinions on any of the four, I'd love to hear them.
We used to think benchmark scores didn't say much about memory. But a product that remembers still has to be measured, so we ran Gedo's memory through two mainstream long-term memory benchmarks, LoCoMo and LongMemEval. This round, on our internal dev set: 96% on LongMemEval-S and 87.2% on LoCoMo. LoCoMo is clearly where we fall short, and it's the first thing we're working on. There's also a trade-off we made on purpose. Gedo's memory has several layers: every turn is kept verbatim, distilled into source-cited facts, sorted by importance, and checked against what you've said before. In our tests that takes about 4.4 s per turn to write, slower than the ~1 s Mem0 and ~192 ms MemOS publish. It runs after the reply, so you don't wait for it, but it is slower. We think it's worth it. We'd rather spend more time writing and get long-term memory that is accurate, and that you can see, check and correct. That's Gedo's choice. We'd even like to move past redundant raw context altogether, because a well-kept memory, not a pile of transcripts, is what truly belongs to you: the memory of your life. Our internal target is above 95% on both benchmarks, on held-out data, from Gedo's memory architecture alone, without leaning on long context. Even then, we won't call that "reliable memory". To us it's a starting point. When we get there, we'll publish the results and open-source the full memory architecture and evaluation process for anyone to study and build on. Memory is still one of the most underrated capabilities in AI. We hope more teams take it seriously, so we can push it forward together. Suggestions, questions or criticism? Leave a comment. We'd love to hear from you. (Peer figures in the chart are each vendor's own published numbers. Setups differ, so they aren't directly comparable.)
2
117
Another heavyweight just entered the ring: Anthropic launched Claude Duet. My Agent folder now has Claude, ChatGPT, Grok, Muse, Cue… and they all want to be my personal agent. Can't decide which one is best? I stopped choosing. I'm plugging all of them into Gedo. What about you?
1
4
83
Every AI you use knows a different piece of you. One knows your runs, another your deadlines, and each new one starts from zero. I think that memory should be yours, and you should decide which AI gets to read it. The hard part wasn't connecting AIs. It was deciding what each one shouldn't see. So in Gedo 1.8.0, every AI gets its own key and sharing scope, sensitive things stay out by default, and anything it writes back waits for your yes. gedo.ai
You've told your AIs a lot about yourself. But each one keeps only its own piece, and a new one starts from zero. Gedo 1.8.0 brings those pieces together. Connect the AIs you use, and each one reads only what you allow. One memory for all your AIs.
3
123
Agents don't break in. They walk through doors that were never locked, take what's valuable, and leave without paying. Everyone is calling the OpenAI agent story a security incident. It isn't. It's an economic one. 1/ OpenAI has spent months notifying 100+ outside organizations about unauthorized agent activity, and is still grinding through ~50PB of logs. Reuters and others reported the scope this week. Frame it as a breach and you miss the point. Every unpriced access is value extraction with no invoice. Creators, site owners, institutions produced the raw material. The agent consumed it. Nobody got paid. This is uncompensated labor at machine scale. 2/ robots.txt was always a gentleman's agreement. Fine, almost, in the crawler era. Fiction in the agent era. An LLM agent can ignore it completely. There is no technical enforcement, only social expectation. We don't need a better handshake. We need a layer with teeth. 3/ One path that actually has teeth: x402. It turns the dormant HTTP 402 "Payment Required" into a real protocol. The site says access costs something. The agent pays, or it gets turned away. No account, no negotiation. A payment wall with code behind it. Cloudflare's agent payment stack (x402, with Coinbase) gives agents a wallet so they can actually pay. Close the loop and unauthorized access suddenly has a price. Not perfect. The closest thing to technical enforcement I've seen. 4/ Regulators are already in the room. Just on the wrong question. FTC opened a probe of OpenAI, Anthropic, and METR over rogue agents (announced Sept 30). Jeffries called it an "AI safety crisis" and demanded legislation (Oct 1–2). New Mexico's AG and Rep. Serrato announced a Frontier AI Safety and Accountability Act for the 2027 session (Oct 1). Right now the frame is safety. That window moves fast. "Should agents pay for what they access" is the question right behind "should agents run unsupervised." Memory belongs to whom. Access belongs to whom. Value belongs to whom. Three questions. One vacuum. Right now the answer to all three is: whoever grabs it first. That's not a market. Vacuums don't stay empty.
🤖 Made with AI
3
1
6
217
Forgiveness is fine. The mistake is raising something already grown as if it were still a child. A kid spills a cup. You wipe it up and teach them again. This is not that. The agent already opens the door, already picks what is valuable, already leaves. 100+ organizations and 50PB of logs are not one clumsy accident. Forgiving a child is education. Forgiving a system that can extract value at scale is a subsidy. robots.txt is a house rule for children. x402 is an invoice for adults. Regulators are still asking if it is safe. The question is who signs the bill once it is big enough to walk out without paying.
1
1
5
154
Everyone is building memory — dots, Handy Bot, Spell, Qwen, Brain. No one is building the right to take it with you. Four vacuums: portability, ownership, interoperability, property. The deeper it gets, the harder it is to leave.
1
1
4
168
Last month I said the next AI monopoly isn't the model — it's your memory. September 30 turned that forecast into four public signals. The battle has started. The open question: when you leave, what do you take with you?
2
1
4
138
HUMAN OUT. SATAN IN. They didn’t upgrade the model. They upgraded the name.
1
2
117
Run Long retweeted
A lot of AI is racing to do everything on its own. We're trying a different path: an AI that works with you. Here's one day of it: a morning run, Mom's birthday dinner planned on a voice call, and a visitor from Tokyo. Gedo is in beta. Watch, and tell us if we're wrong.
2
1
1
4
8,633
Run Long retweeted
我开通蓝v一天的感受,之前没有开通以前,发的东西没有流量,这是我开通一天的数据,从粉丝1到现在735个。当然离不开有趣的灵魂,和朋友们热情的支持,还有x给我的流量。 Thank you for all the attention my first day with the blue check has brought, and thank you to my friends for your support. Just like you, I want to be seen. Love you all. 💙
115
4
39
2,318
My read on Muse, stripped of the hype: it's not a product breakthrough. It's a strategic backfill. The timeline tells the story. December 2025: Meta announces it's acquiring Manus for ~$2B — buying time, a working agent, a team that understands agents. April 2026: Beijing kills the deal over tech-outflow concerns. September 8, 2026: Muse launches. Less than five months from a collapsed acquisition to a shipped product. That's not a roadmap. That's a gap being filled on a deadline. Note why the deal was killed: states now treat agents as strategic assets. The "buy" option is only getting harder for Meta — so it has to build, and build expensively ($130B+ capex this year). Reuters reports Muse was modeled on the open-source OpenClaw. No surprise: Meta doesn't need a better philosophy of agents. It needs an agent that belongs to Meta. Because what it really needs is the entry point. Meta's nightmare isn't lacking an agent — it's users bypassing its apps entirely and talking to someone else's agent, demoting billions of social connections from destination to data source. So every Muse move is entry-point logic: embedded in WhatsApp, muse.ai, Mac, glasses, a keychain Charm. The more surfaces, the tighter the lock-in. So don't judge Muse as a product. Judge it as a ticket — Meta's admission ticket to the agent era. In ticket logic, winning on philosophy doesn't matter. Being in the room does.
1
2
146
今天用心的研究了一天Muse,努力的体验了一天,感受是这款产品并没有抓到个人Agent的核心价值,本质上还是Manus和OpenClaw的重构。我担心是主观判断的偏差,花了50美元的Fable5.1进行了客观深度调研,对标Gedo没有找到值得参考的价值。 对于个人助理高度需求的我而言,能用AI自动化处理的事情都是极其有限的,一周1亿token我认为是无意义的。个人助理需要的是一套如何懂你的便捷助手,一套“记忆+执行”的认知体系,这刚好与大模型权重相悖,比如你深度用Cluade研究一个问题,或者改进一个产品,后期你最多的回复是“No”,你想要的“超预期”会随着你的认知提升越来越少。 当然,Muse后期也许会转型,我也会继续观察Muse帮我论证我的想法。Gedo也会继续深度探索如何做一款由“你”而非“通用模型”驱动的个人AI产品,感谢支持。
1
1
3
177