Antirez frames DS4 not as a Redis replacement but as a fundamentally different take on the data structure server concept. Having created Redis in 2009 and led it for over a decade, he is exploring what the category looks like when designed from scratch with full knowledge of Redis's accumulated trade-offs — single-threaded ceilings, bolted-on modules, and cluster complexity.
The editorial emphasizes that DS4 represents antirez's answer to a question only he can ask: 'if you could start over knowing everything Redis taught you, what would you build?' It positions the project as taking Redis's hindsight lessons seriously rather than patching over existing limitations.
The editorial argues that Redis's single-threaded event loop was simultaneously its defining advantage (no locking, predictable latency) and the constraint that prevented it from saturating modern multi-core hardware. The module system and cluster mode were identified as later additions that felt bolted on and introduced operational complexity antirez himself was publicly ambivalent about.
Submitted the DS4 blog post to Hacker News where it garnered 330 points and 132 comments, signaling strong community appetite for antirez's post-Redis work. The level of engagement reflects how deeply Redis is embedded in modern developer stacks and how much weight the community places on antirez's design instincts.
Salvatore Sanfilippo — known universally as antirez — published a detailed blog post titled "A few words on DS4" on his personal site (antirez.com/news/165), giving the developer community its first substantive look at his post-Redis project. The post hit 330 points on Hacker News, reflecting the intense interest the community has in whatever the creator of Redis builds next.
Antirez stepped back from Redis development in 2020 after leading the project for over a decade. Redis, now managed by Redis Ltd. (and forked as Valkey by the Linux Foundation after the 2024 license change), remains one of the most widely deployed pieces of infrastructure software in the world. DS4 represents antirez's answer to a question only he is uniquely positioned to ask: if you could start over knowing everything Redis taught you, what would you build?
The post lays out DS4's core philosophy and early technical direction, positioning it not as a Redis replacement but as a fundamentally different take on the data structure server concept.
### The Redis hindsight advantage
Redis was born in 2009 as a practical hack — antirez needed a fast data structure server for his startup's real-time analytics. Its genius was in what it *didn't* do: no query language, no schema, no complex replication protocol (at first). That simplicity made it the default "fast thing" in nearly every modern stack.
But Redis also accumulated design decisions that are difficult to reverse. The single-threaded event loop — Redis's defining characteristic — became both its greatest strength (no locking, predictable latency) and its ceiling (you can't saturate modern hardware). The module system, added later, felt bolted on. Cluster mode introduced operational complexity that antirez himself was publicly ambivalent about.
DS4 takes these lessons seriously. Rather than patching Redis's limitations, antirez is exploring what a data structure server looks like when designed for the hardware and usage patterns of 2026, not 2009. The project appears to rethink core assumptions about threading, persistence, and the boundary between data structures and computation.
### Design philosophy: smaller, sharper, opinionated
Antirez has always been a minimalist. Redis succeeded partly because he was willing to say "no" to features that didn't fit the mental model. DS4 continues this tradition but with different constraints. Where Redis optimized for the widest possible set of use cases (cache, message broker, session store, leaderboard, rate limiter), DS4 appears to be more opinionated about what it wants to be.
This matters because the data infrastructure landscape has fractured since Redis's early days. We now have specialized tools for every niche: DragonflyDB for multi-threaded Redis-compatible workloads, KeyDB for the same, Garnet from Microsoft for .NET-native performance, and Valkey carrying the open-source Redis torch. DS4 isn't competing with any of these directly — it's asking whether the Redis API contract itself is the right abstraction.
### The creator-revisits-creation pattern
Antirez joins a notable pattern of original creators returning to rethink their most famous work. Rich Hickey built Datomic after Clojure. Ryan Dahl built Deno after Node.js. Brendan Eich keeps trying to fix JavaScript's sins through various means. These "second system" projects don't always win market share, but they almost always produce ideas that the ecosystem eventually absorbs.
The HN discussion (330 points suggests strong engagement) likely reflects this dynamic: developers follow antirez not because they'll immediately adopt DS4, but because his design instincts have an outsized track record of being right about what developers actually need.
### Don't migrate anything (obviously)
DS4 is in its early public disclosure phase. There is no production deployment story here and won't be for some time. If you're running Redis, Valkey, DragonflyDB, or any other in-memory store, nothing changes today.
### Do study the design decisions
The value for practitioners right now is architectural. Antirez's blog posts have historically been masterclasses in systems design thinking — the "why" behind technical choices. Reading how he's rethinking data structure servers is directly applicable to how you design your own systems, even if you never use DS4.
Specific areas to watch: - Threading model: How DS4 handles concurrency will signal whether the single-threaded-with-io-threads approach (Redis 6+) was an evolutionary dead end or just needed refinement. - Persistence strategy: Redis's RDB/AOF duality has been a source of operational pain for years. DS4's approach here could influence the broader ecosystem. - API surface: Whether DS4 maintains Redis protocol compatibility or breaks clean tells you a lot about where antirez thinks the real value of Redis lives — in the wire protocol or in the data structure semantics.
### The Valkey/Redis context
DS4 also matters as a data point in the post-license-change Redis ecosystem. With Valkey under the Linux Foundation carrying forward Redis compatibility, and Redis Ltd. pursuing its commercial strategy, antirez building something entirely new suggests he sees more value in rethinking the problem than in fighting over the existing solution. That's a meaningful signal from the person who understands Redis's architecture better than anyone alive.
DS4 is a long game. Antirez is a deliberate builder — Redis itself took years to reach the level of polish that made it ubiquitous. The smart move for the community is to follow the development closely, engage with the design discussions, and extract the architectural insights even if DS4 itself never reaches Redis-scale adoption. The best-case scenario is that DS4's ideas improve every data structure server in the ecosystem. Given antirez's track record, that's not a bet against the odds.
I'm very curious where we will saturate the curve on "enough" intelligence for coding. At some point, you can let a less smart model hammer at a problem for longer and get to the same result, and as long as you are not involved it comes to the same thing. I feel like DeepSeek V4 Pro i
Great to find this narrow focused thing:> We support the following backends: Metal is our primary target. Starting from MacBooks with 96GB of RAM. NVIDIA CUDA with special care for the DGX Spark. AMD ROCm is only supported in the rocm branch. It is kept separate from main since I (antirez) don&#x
It’s shocking how close this feels to claude, obviously it's much slower, but I don’t know that it’s significantly dumber. Interestingly the imatrix quantization seems to be better than whatever quant the zdr inference backends on open router are using. It was self aware enough yesterday to rea
I've been using the Q4 version on my Mac Studio over my local network and it's been good. Indeed, I had the first ever experience where I was playing with it alongside my various other agents and forgot it was a local model as it was doing such a good job.I do wonder, though, if another ag
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
DwarfStar4 is a small LLM inference runtime that can run DeepSeek 4. The blog post implies that it currently requires 96GB of VRAM.For others who are lacking context :-)