Styrir Prime

Working on tools to advance an interplanetary agenda.

[Styrir Prime @StyrirPrime]

We released Rúnir. https://t.co/vg6nqBovat

Rúnir is a discrete memory service for AI agents. Agents call it via HTTP once per turn: before the model replies, the service recalls the essential bits; afterward, it archives what deserves retention; finally, it tidies up. The rule is simple: just in time, just enough. That continuity isn’t a dump of your entire interaction history; it’s that narrow, curated fragment of retained knowledge worth placing into the prompt.

Memory here, then, is understood as living sense, not log scrapbooking. You save lived instances as semiotes (what was said and what was resolved), which feed semiosis (continuous interpretation of changes of context), which in turn modulate hexis (appropriate dispositional and affective stance in retrieval), while more permanent patterns coalesce into noema (stable sense-particles that survive the life of a conversation), and syndesis stitches the chain across time so future agents can still pick up the story. All writes are arbitrated (create/merge/supersede/skip) to allow facts to change without losing lineage, with “soft-inactivate” the path of least resistance and “hard-delete” an explicit choice. Retrieval is a hybrid mix (vector search, full text, recency, entity lookup), after which weakly-matched items are filtered to protect the token budget.

What we want to be distinctive here is separating the concerns: intelligence lives within Rúnir’s service layer, while Claude, Codex, Pi, Grok Build, OpenClaw, etc. act as thin clients. No mandated framework, no “session-starts-as-novel-as-memory” fiction, no expectation the model fabricates continuity from thin air when the service returns nothing. Ambient hooks support the ordinary “remember that I like …” style capture and recall; when you do want to know, explicit tools and deep surfaces (e.g., /search with citations) give straightforward answers even when the memory doesn’t know. Rúnir is for agent builders looking for continuity that compounds: less copy-pasting, fewer repeat explanations, more retained sense.

[Styrir Prime @StyrirPrime]

@herdrdev @BessemerVP @ycombinator @e2mindset @tobi @dok2001 @gorkem Congratulations. A well-deserved round.

Been using herdr since March.

[herdr @herdrdev]

Herdr has raised a $6M seed round led by @BessemerVP, with @ycombinator, @e2mindset, and angels including @tobi, @dok2001 , and @gorkem.

Oh, and we're also hiring 👇

https://t.co/8VXU3Nwehh

[Styrir Prime @StyrirPrime]

@crutchcorn Maybe it's time agents started writing code for disk efficiency, token efficiency, and for agent reads rather than human reads. Are you always reading your code, because most people don't anymore.

[Styrir Prime @StyrirPrime]

@theo I have too many deterministic evals and verifications in my workflows/workgraphs for my models to go off point this radically. I would suggest you do the same, regardless of which models you're using.

[Styrir Prime @StyrirPrime]

@forgebitz Use Grok build with CLI proxy, then you can use whatever you want. Just have your agent add your models to the config file in Grok Build, and you're good to go.

[Styrir Prime @StyrirPrime]

Understood. I suppose there are always going to be some friction when you've got people who don't really have a programming background. They vibe code everything, might have a YouTube channel or a following, and get to these places without having any real engineering foundation. If someone tries to pierce their influencer veil with legitimate concerns, they get offended, just like this guy did (or so it seemed). I'm still going to keep giving people recommendations and suggestions, especially where best practices are concerned.

[Tibo @thsottiaux]

To calibrate you all on which reasoning effort to use for Astra, know that GPT-6 Astra on low performs better than GPT-5.6 Sol on high.

If you were using high reasoning efforts with Sol and were happy, I suggest you move down to low or medium for Astra.

[Styrir Prime @StyrirPrime]

@ChickenOnAStrip @emanueledpt No, I was giving him a suggestion.

[Styrir Prime @StyrirPrime]

So far, what we're seeing is that Astra finishes most tasks in our typical workflows, which have a lot of deterministic tooling between each step of the workflow, are finishing in about half the time versus Fable. Token cost is also a lot lower - don't have a large enough sample size yet to give you an accurate reading, but I'd say it's at least 20-40% cheaper on token spend - some tasks, even larger gap.

[Styrir Prime @StyrirPrime]

@Zerkoto @jp54362 Hasn't been the case on my end. Make sure your context window isn't getting bloated from too many skills and MCPs.

[Styrir Prime @StyrirPrime]

Mostly, what I'm seeing regarding criticisms of Astra is that people are burning through tokens by setting the model to higher effort levels than they need to. There have been several people on OpenAI's team who've explicitly stated that, for most coding work, you can set the effort to medium.

We're bouncing back and forth between medium and high, so our workflow will use Astra High for bigger planning scope, and then we'll use medium effort for the coding. It works great for us, much better than Fable. Tasks get done in half the time and with a lot less token spend.

Vibe coders shouldn't be in a position to judge anything. They should talk less and listen more.

[Styrir Prime @StyrirPrime]

@iuditg That's the opposite of our experience, but we're not vibe coding things, and we run our models through deterministic gates at every step in the workflow.

[Styrir Prime @StyrirPrime]

@emanueledpt I do. I'm actually pretty goddamn funny.

This has nothing to do with best practices, however.

[Styrir Prime @StyrirPrime]

So you're saying that it does abstractions well, but you don't like the output of the code. That’s not what I’d call a correctness bug. It's more of an idiom or style, yes?

Run rustfmt + clippy (pedantic on) as a hook after every model edit in your workflow, feed any failures and warnings back as the next prompt, require a clean pass before accepting the diff. This is going to be the single biggest quality improvement you can make, and it’s basically deterministic.

Also, put 2-3 examples of your idiomatic Rust in context (or a CLAUDE.md/style doc) rather than relying on the base model's Rust "taste".

If you're doing any agent coding at scale, use a clippy.toml/lint-level config checked into the repo so the loop's pass/fail bar is fixed and reproducible rather than vibes-based.

[Styrir Prime @StyrirPrime]

@emanueledpt I doubt that you need that much reasoning on a standard repo audit. I wouldn't be surprised if you burned through one of your 5-hour session limits doing this.

[Styrir Prime @StyrirPrime]

For most of the things that we're doing, yes, Astra completes the tasks faster with fewer lines of code and in a third or half the time.

We can test and measure these things for our evaluations because we have standard workflows in our system with a lot of deterministic gating between each step. That means we can measure what each model is doing from step to step on these workflows.

[Styrir Prime @StyrirPrime]

That's the big problem: most people are using it on high because they use the older models on high, and you should likely be using it on medium for most things. Fable and Astra are very advanced models. I think, for the majority of the work that most people are going to be doing, you're not going to need anything above medium for average tasks.

[Styrir Prime @StyrirPrime]

@buildwithrajath Try it on medium effort, and you probably won't miss the higher settings for the majority of the work that you're doing.

[Styrir Prime @StyrirPrime]

@Im_IrushiK You probably don't need to be using high effort. Try it on medium.