Gedo Labs (@gedolabs) header
Gedo Labs (@gedolabs) profile picture

The Life OS that knows you. Other AIs know your questions. Gedo knows yours.

Singapore
Joined July 2026
On November 19, the results of the second CSIG AML Memory Challenge were announced. AML (Agent Memory Leaderboard) fixes the answering and judging models, so entries differ only in their memory layer; this cycle completed more than 400 formal evaluations. Gedo Memory placed 3rd on the Commercial Products leaderboard for Text Memory (62.28). After taking part, we wrote up our design and what this cycle showed us as a research note: "Designed to remember your life: the system design and trade-offs of Gedo Memory." The two #1 systems and our own happen to be three different approaches. ① Verbatim retrieval Resurface, #1 on the Open-Source Methods leaderboard (61.22). No model calls. Every message is stored as is, retrieved and handed to the answering model. Near-zero write cost and nothing lost; updates and reasoning are left to the answering model. ② Fact extraction MemoraX, #1 on the Commercial Products leaderboard (69.26). Built around extraction and add, update or delete decisions at write time, so reads stay light. Done well, it pays off most, and that takes strong extraction and update handling. ③ Two layers Gedo Memory, #3 on the Commercial Products leaderboard (62.28). The verbatim record is kept without loss as the fallback, and only the important facts are extracted as the index, accepting misses. That is why both the product and our entry run extraction on a small model. Our internal pilot (LongMemEval-S) Verbatim retrieval reaches about 87%, already close to putting the whole history in context. The two layers together reach about 93%, matching the whole history with about 5% of its input. On LoCoMo, verbatim retrieval does better. What we took away · The verbatim record is the floor. Extraction alone is the riskiest option. · Extraction earns its keep above that floor: time anchoring, updates and provenance. · The per-capability scores also show what our choices cost: the widest gaps are in epistemic safety and privacy and in personalization, which map to how much context we hand over and to passive retrieval. · The verbatim record can be found, but it can hardly be seen. We want your memory to be fully visible, so high-quality extraction is our core research direction. Thanks to @AgentMemoryL and CSIG for the evaluation work. Full note: the four approaches, the cost of each design decision and what comes next gedo.ai/labs/gedo-memory-des… #AgentMemoryLeaderboard
1
3
92
Gedo Labs retweeted
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
Honored to place 3rd on the AML Cycle 2 Commercial Products leaderboard (Text Memory). Thank you @AgentMemoryL and CSIG. Memory is what lets an agent truly know the people it works for, and an open, rigorous benchmark like this helps the whole field measure it honestly.
🏆 AML Cycle 2 results are live! 💥 • 1M+ cross-platform impressions • 10M+ cumulative website clicks • 402 task submissions • Three tracks: text, code and multimodal memory This cycle brought together teams from across the global AI community, including teams from Alibaba Cloud DMS, Shopee, OPPO, KT and OceanBase—alongside university researchers, memory startups and independent developers. Explore the full results: agentmemoryleaderboard.ai Highlights and evaluation details below 🧵
1
1
4
131
🏆 AML Cycle 2 results are live! 💥 • 1M+ cross-platform impressions • 10M+ cumulative website clicks • 402 task submissions • Three tracks: text, code and multimodal memory This cycle brought together teams from across the global AI community, including teams from Alibaba Cloud DMS, Shopee, OPPO, KT and OceanBase—alongside university researchers, memory startups and independent developers. Explore the full results: agentmemoryleaderboard.ai Highlights and evaluation details below 🧵
7
7
7
22
44,645
Gedo Labs retweeted
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
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
1
3
37,132
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
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.)
6
1
4
329
What we'd most like your input on: how you'd test memory beyond these two benchmarks, what "reliable memory" should mean, and where an AI should draw the line on what it remembers. Reply here or DM us.
1
18
On the ~4.4 s write time: it runs after the reply is sent, so you don't wait for it. The time goes into keeping your words, extracting facts with sources, catching values that changed, and sending anything uncertain to an inbox for you to confirm. That's where we'd rather spend it.
12
How Gedo memory works, in one line: facts are the index, the original words are the answer. Each turn is stored verbatim, distilled into short facts that point back to their source, and sorted into core, active and cold tiers. When you ask something, we find the facts, then give the model the exact words they came from.
12
How we tested: Gedo answers from its full memory layer (extracted memories plus the original conversation), with Claude Sonnet 5.5 as the answer model, on an internal dev set. LongMemEval uses its official judge; LoCoMo uses an LLM judge with Mem0's public prompt. Peer numbers are each vendor's own published figures, with different models and setups, so read the chart as context, not a ranking.
1
64
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.
1
1
5
33,878
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
Muse asks: how do things get done "without you?"Brief it, approve, close the app — it keeps going. The better it works, the less you need to show up. That's not a flaw in the design. That is the design: an agent in your place. We ask a different question: "what can you get done?"Not removal — expansion. An agent that grows with you can't treat you as a bottleneck to route around; it has to know you. Your memory, visible and correctable. Your goals, broken into today's one step. Every round it knows you better, and you get stronger. I understand why Meta builds Muse. It's a distribution play: own the entry point, keep the human inside their world. Worth respecting. Not worth copying. Two ideas of a personal agent. We picked ours. Which question do you want yours to ask?
Muse and Gedo are both personal agents. The difference is who sits at the center. Muse puts the AI there: it reasons and acts on its own computer, then brings you results. Gedo puts you there: it works from your memory, which you can see and fix, and moves your goals forward.
3
136
You can call Gedo now, on web, iOS and Android. Tap the phone in any conversation and talk it through: on a walk, at the stove, before you drive off. Cut in whenever you like. What it drafts lands on your screen mid-call. Hang up, and the whole call is still in the chat.
2
1
3
312