DV retweeted
#PhysicalAI is moving from lab demos to deployed systems. Robots that perceive, reason, and act in the real world.
But what’s working, what’s hype, and what does #Europe need to win?
Our CTO Simone Totaro will talk about it Friday 20th November in #Zürich at DARE Campus.
DV retweeted
GitHub 📉. @rwx 📈. @brilliantmade 📈.
The hosted RWX MCP server is now available. Allow your agent to read run/task results, logs, artifacts, and documentation. rwx.com/docs/mcp
Great to see the Fujitsu team putting together a hands-on tutorial on building reliable workflows with DBOS and Fujitsu Enterprise Postgres.
One highlight is how DBOS users, especially those running sovereign or on-premises deployments, can benefit from Fujitsu Enterprise Postgres' Transparent Data Encryption to meet their security requirements.
Search should be like electricity and water for AI: ubiquitous, affordable, and always available.
That’s what we’re building at Octen AI.
Thanks to @carlfranzen and @VentureBeat team for the thoughtful coverage!
Ex-Alibaba engineer's new startup Octen pairs 10-millisecond AI search with a $20 unlimited plan
venturebeat.com/ai/ex-alibab…
10ms. Unlimited for $20/month. Now multimodal.
Today, @OctenAI is launching:
1. Octen Pulse: Open-source search at the speed of a single token.
2. Unlimited Search: $20/month. No more counting API calls.
3. Multimodal Search: Text and images. Platform.
Humans search with Google. Machines search with Octen.
DV retweeted
Ex-Alibaba engineer's new startup Octen pairs 10-millisecond AI search with a $20 unlimited plan
venturebeat.com/ai/ex-alibab…
DV retweeted
What led our CEO, Elisa Seghetti, into the space of #WorldModels and why she decided to start a company within it? Watch the entire conversation of the Atoms are Back panel at #Wave by @vento_ventures on our Youtube channel. Link below.
DV retweeted
If a cheap model already solves your task, adding a Skill is a weak test. The better question: can a verified procedure from a stronger model help it complete work it struggles with? @zep_ai
#rust won’t rust 😎
Busy release day at DBOS! We just released DBOS Rust 0.6.
This is a major update, bringing a bunch of new capabilities to Rust:
- Durable streams and scheduled workflows
- Debouncing and workflow patching
- Workflow rewind
- Portable and custom serialization
- Support for the smol and async-global-executor runtimes
We've also significantly expanded our Rust documentation, with a dedicated section on the DBOS docs site, including programming guides, tutorials, and API references. Every feature is now documented on docs.rs as well.
Note that 0.6 introduces breaking changes to both the API and system database for users upgrading from 0.5.
Full release notes: github.com/dbos-inc/dbos-tra…
DV retweeted
Whoever knows Cesc knows it’s intelligence is uniquely matched by increadible kindness and humbles.
Grateful to be building alongside him and Simone 🙌🏻
🎬 Field Notes: world models for robotics?
Great chat with Cesc Cunillera from @saturndynamics about how world models can work for robots in production, in industrial factories. 👇
DV retweeted
🎬 Field Notes: world models for robotics?
Great chat with Cesc Cunillera from @saturndynamics about how world models can work for robots in production, in industrial factories. 👇
DV retweeted
Following Wave by @vento_ventures in Turin, our CEO Elisa Seghetti sat down with @repubblica to unpack how #SaturnDynamics uses physics-faithful world models to accelerate real-world #robotics deployment. Italian article by @dariodelia74 in the first comment 👇
DV retweeted
Octen Search's search cost is about $9 per 1,000 tasks, from roughly 6 to 12 searches per task at $1 per 1k queries (an 80% discount from their standard $5/1k query pricing).
Model cost is about $49 per 1,000 tasks, mid-pack, because Octen's highlight payloads are mid-sized at roughly 7k tokens per search.
At $0.058 per task and an index of 77, Octen sits on the Pareto frontier for Artificial Analysis Search Index vs. Cost per Task, between TinyFish Search ($0.035, index 71) and Perplexity Search (medium) ($0.091, index 80).
DV retweeted
Octen Search's 16.9 seconds per task is below the model-only baseline of 22.6 seconds: searches return in about 0.21 seconds on average.
Octen sits on the Pareto frontier for Artificial Analysis Search Index vs. Time per Task.
DV retweeted
Octen Search debuts third on the Artificial Analysis Search Index with a score of 77, achieving the fastest Time per Task at 16.9 seconds and one of the lowest Costs per Task
We benchmarked Octen Search from @OctenAI using our search agent methodology: a fixed base model running inside our open-source agent harness, Stirrup, with tools for searching and fetching pages from the web. Only the provider behind the search tool changes.
Key results of Octen Search with highlights enabled:
➤ Strong Search Index performance, led by BrowseComp: Octen Search scores 77 on the Artificial Analysis Search Index, placing third behind Perplexity Search (medium, 80) and Perplexity Search (high, 79), and level with Perplexity Search (low, 77). Octen performs particularly well on BrowseComp, scoring 86, just one point behind the leader. It also scores within one point of the leaders on DeepSearchQA at 80, while its AA-Omniscience score of 66 trails Firecrawl Search (73) and Perplexity Search (medium, 72)
➤ Fastest search across providers: Octen Search averages 0.2 seconds per query, 5x faster than the Perplexity Search variants at 1.0 to 1.2 seconds. Its Time per Task is 16.9 seconds, the lowest of any provider result, against a board range of 16.9 to 62.9 seconds
➤ Low Cost per Task: Octen Search costs $0.058 per Search Index task at a price of $1 per 1k search queries. Only TinyFish Search, which offers search for free, has a lower Cost per Task. For comparison, Perplexity Search (medium) costs $0.091 per task
DV retweeted
I’m excited to announce that @GoEntire now supports running CI through @RWX. Rebuilt from the foundation up for how agents work:
➡️ RWX: a graph-based execution model with automatic caching. Far less compute needed per loop.
➡️ Entire: Trails replace the pull request, keeping context, code, and verification in one place, even across repos and agent sessions.
rwx.com/blog/entire-integrat…
Native integrations for CI build another layer of developer choice and control into the fastest Git forge on Earth.
Today, we’re excited to announce our partnership with @rwx , an agent-native CI platform now built directly into the Entire plugins. Try it out!
entire.com/blog/ci-with-rwx-…
DV retweeted
.@zep_ai introduced Zep Agent Skills, and it’s really worth your attention.
It's a procedural memory system that turns verified, successful AI agent runs into reusable Skills.
The idea of procedural memory isn't new, but Zep is trying to turn it into a managed service, with a few technically interesting details ↓
First, here's how it works:
1. Tracing: Zep records the agent's actions, tool calls, and outcomes as a Trace. It can also import Traces from Braintrust and LangSmith.
2. Verification: Checks whether the task actually succeeded using independent evidence, not just the agent's claims.
3. Compilation: A Reflector proposes reusable instructions, while a Curator checks them against the evidence. Each Skill field links back to specific Trace events.
4. Admission: Safety is a very important point. Security scans and quality gates determine whether a new Skill version becomes available.
5. Retrieval: Agents find relevant Skills through lexical search, load instructions when needed, and record outcomes that influence future rankings.
Then, there are a few features that are particularly interesting and make Zep's system distinct:
• Skills aren't tied to one model. A stronger model can solve a task once, and a weaker model can reuse the procedure without retraining.
• Zep uses the open Agent Skills format. Each Skill is packaged as a SKILL.md file with YAML frontmatter and optional files, so compatible harnesses like Claude Code and Codex can load it.
• Every Skill field needs evidence. The compiler links instructions to specific Trace events, making it possible to inspect where they came from. It can also patch individual fields instead of rewriting the whole Skill.
• Skills have a lifecycle. They’re versioned, checked before activation, and ranked partly by verified outcomes. A Skill can even start from one verified success, but remains provisional until it gathers more evidence.
The results are also fascinating:
- Skills learned from Opus 5.5 improved three weaker models' pass rates on the same tasks by 4.5–18.8 percentage points across 3 benchmarks.
- On EnterpriseOps-Gym, all three weaker models using Skills at low reasoning effort beat max effort without them.
Most gains came from tasks the stronger model had already solved.
Try Zep Agent Skills yourself here, you may find something interesting: getzep.com/zep-agent-skills/…




























