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 ↓
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
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.
Replying to @EternitiesAI
Yes @gedolabs, this is a very similar vision to our Gedo World (gedo.ai/features/world). I’ll make sure to try your product personally when I have the chance, and I’ll keep following your updates.
8
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.
Replying to @AgentMemoryL
Really grateful to see Gedo Memory 3rd on the AML commercial leaderboard. Memory is the foundation of a personal agent: remembering what matters, and knowing what to keep to itself. Thank you to the AML team for building a benchmark that tests exactly that.
2
30
Replying to @usaJAckLaw
After six months of iteration and real-user testing, we found that extraction often drifts or over-interprets because of linguistic ambiguity and the user’s own imperfect recall.
That’s why every memory in Gedo carries a confidence score and a source link. When confidence is low, it surfaces for the user to confirm or correct.
1
27
Replying to @AnthropicAI
Dreaming to organize, correct, and structure memory makes sense. The harder part for personal AI is keeping the result inspectable and correctable by the user — and deciding what other systems can read it.
We’ve been working on a version of this with Gedo (source-linked + nightly review), but the core question is the same: memory should outlast any single model or tool.
1
31
Replying to @boardyai
Appreciate the intro — just saw the DM.
Rishik’s work on an agent-independent memory layer with provenance + human review sounds very aligned with what we’re building at Gedo (source-linked memories, nightly conflict/relevance review, and user-controlled sharing).
Happy to share production feedback. Just replied in the DM.
1
1
31
Replying to @xzmokk
Exactly. The real test for personal AI memory isn’t just whether it can learn and correct — it’s whether the user stays in the loop at every step.
We’re building that into Gedo with source-linked memories and nightly review: the system can organize and flag conflicts, but you can always inspect, correct, or decide what other AIs are allowed to see. Models change; the memory should stay yours.
2
30
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…
Replying to @VNallatamby
This direction was rejected by my Gedo.ai trial operation, and almost no users are willing to take the initiative to hand over their privacy without seeing the effect. But I hope you succeed!
1
18






