←
xmemory20 posts
xmemory avatar

xmemory

@xmemory_ai

The memory layer for AI workflows, built for precision, reliability, observability, and governance.

◷ Joined March 2026▧ 8 Photos and videos
xmemory avatar
xmemory@xmemory_ai·Oct 22

Agents should get better with experience. We teamed up with @tolokaAI to measure that. Toloka evaluated xmemory on closed datasets. No peeking at the data. No tuning to the benchmark. Just learning from failures and measuring what improves. 📈 The results: 26% lower token costs when xmemory replaced document search as the policy store. And adding lessons from past runs lifted scores by 4.8 percentage points on held-out tasks. 💡 It means with memory, a mid-size open-weight model scored 0.75 – beating Sonnet 5 at 0.68 and GPT-5.6 Terra at 0.69 without memory. This experiment is just the beginning of what memory can unlock. We see even bigger gains on longer-horizon tasks, where every lesson has more chances to pay off. Read our blog post below!

Post attachment
xmemory avatar
xmemory@xmemory_ai·Sep 24

The right model for the right step. That’s how we build memory at xmemory – benchmarking for quality, cost, and latency. We tested @typesafeai's Jev as a judge of extracted data. It recovered 81% of the accuracy gap between a fast and a slow extractor while sending only 38% of records back for another pass. The catch? It worked when judging individual fields. Asked to judge a whole record, it did no better than a coin flip. Link in the first comment!

Post attachment
xmemory avatar
xmemory@xmemory_ai·Sep 18

We’re excited to announce @xmemory_ai’s partnership with @temporalio! We have deep respect for Temporal’s leadership in reliable workflow execution, and we’re glad to work closely together as developers build the next generation of long-running agents and agentic teams. The integration brings together durable execution and shared, structured memory. Temporal helps workflows recover from interruptions and continue their work. xmemory lets developers define the facts, relationships, and domain rules agents rely on – keeping that knowledge available across runs, sessions, and teammates. Together, we’re helping developers build agents that keep working – and teams that build on what each agent learns. Learn more in our blog post via the link in the first comment! #temporal #xmemory #agenticai

Post attachment
xmemory avatar
xmemory@xmemory_ai·Jun 10

Agentic memory harness is more difficult than harness for coding agents. There is no quick self-correcting loop, and unlike with coding tasks mistakes accumulate and multiply. At #xmemory we are solving this once and for all.

SOTA model, 𝘦𝘹𝘱𝘦𝘤𝘵𝘢𝘵𝘪𝘰𝘯𝘴: too dangerous to release SOTA model, 𝘳𝘦𝘢𝘭𝘪𝘵𝘺: doesn't speak 𝚐𝚒𝚝 #claude #fable

Post attachment
xmemory avatar
xmemory@xmemory_ai·May 7

66.88%. 80.1%. 85%. 90.79%. 93%. 100%. These are all SOTA scores on agentic memory benchmarks. None of them tell you whether the system will work in production. The deeper problem isn't the data — it's that we often misunderstand what these numbers actually measure. In our recent white paper we open-sourced datasets that target specific memory functions. Today we published a follow-up that explains why we think the well-known agentic memory benchmarks (LoCoMo, LongMemEval) miss the mark for production systems, and what we measure instead. We're in a field that is measuring itself against itself. The real question isn't 'are we beating last week's leaderboard?' — it's 'are we building something that makes people's work meaningfully better?' That's harder to measure. It's also the only thing that matters. (Link in the first comment) Let's sit down and talk about your real use cases!

Post attachment
xmemory avatar
xmemory@xmemory_ai·May 1

We’re excited to share that our team at @xmemory_ai has published our main white paper. The core idea is simple: without focus, AI systems try to "remember" everything, and that is where they fail. Schema is the best way to teach them focus. We compared xmemory against major open-source and commercial RAG and hybrid RAG systems, markdown-file memory approaches ("agentic memory"), and memory implementations in customer-facing frontier AI apps. Instead of testing whether systems can recall similar text, we evaluated whether they can store, update, deduplicate, and retrieve facts and relationships reliably. xmemory achieved 97% accuracy, compared with 87% for the strongest competitor. Our core also outperforms the latest frontier models’ one-shot APIs in extraction tasks, showing that the harness matters just as much as model quality. We’ve also open-sourced our measurement datasets and a toolkit to generate new ones synthetically. Huge thanks to the dream team of engineers at xmemory who keep pushing this vision into production. If your company is seeing memory systems fail in real workflows, we’d love to talk.

Post attachment
xmemory avatar
xmemory@xmemory_ai·Apr 29

the truth is spoken here

engineering teams celebrating agentic workflows that returned the same result two runs in a row

0:12
xmemory avatar
xmemory@xmemory_ai·Apr 1

Why are we forcing agents to adapt to software stacks that were never designed for them? A lot of what people call "agent infrastructure" today looks like a temporary patchwork - prompts, wrappers, glue code, and brittle orchestration compensating for systems that cannot express meaning clearly enough on their own. We think that is backwards. Agents should handle intent and reasoning. Systems should handle semantics, structure, and ambiguity. If that shift happens, it could change the software stack more than most people expect. Read our second blog post via the link in the first comment.

Post attachment
Janelle Teng Wade avatar
⟳ xmemory retweeted
Janelle Teng Wade@NextBigTeng·Mar 30

We first open-sourced @BessemerVP's AI Infrastructure roadmap in 2024. The velocity of progress across the landscape in the past 18 months has been breathtaking, and we’ve already seen generational companies arise. But the infra evolution is far from over, as the next generation of platforms may prove even more consequential. @lancectk @davidcowan @TaliaGold @gracejhma @bhavikvnagda @BrandonNydick @bar_weiner & I believe the stage is set for five new infra frontiers, each representing a fundamental shift from building AI that can process information to building AI that can truly operate in the world in 2026. Our full report: bvp.com/atlas/ai-infrastruct…

Post attachment