Twitxy Saved

june

Verified account This account is verified. Learn more

@arjunkmrm

exploring composable, self-improving agent harness tardigrade.sh. prev smithery.ai (acquired)
Singapore Joined June 2015
1.3K Following 2K Followers 2.5K Posts

Latest posts 5

Pinned post

june Verified account @arjunkmrm
Introducing Tardigrade! 🐛

A framework to build your agent harness as components over an immutable event log (think react for harness).

Each component is a typed state machine over the event log, using @EffectTS_.

Add to your effect code, run locally or self-host as durable objects with cloudflare or celld.

tardigrade.sh/docs/why
tardigrade.sh Why Tardigrade? | Tardigrade Learn how Tardigrade builds composable agent harnesses from components over an immutable event log.
43 replies 49 reposts 625 likes 83K views View on X
june Verified account @arjunkmrm
I shared a little about my mental model of durable objects here (linked). In summary, it implements the actor model of computing. A great example of which is how humans in a distributed network function in the real world! Maybe I should write about this as a complete blog.

It’s one of those things that took a while for me to appreciate, but once it clicked, made me question why this design isn’t universal.

It isn’t just about the storage or logs, but the fact that it’s also colocated with compute. And there’s consensus in operations that change state.

x.com/arjunkmrm/stat…
David Cramer Verified account @zeeg 18h
designing this thing that will work w/ Durable Objects and the more I try to use them the less I understand about them

imo the complexity of DOs (and Workers) is probably the biggest weakness of CF's platform. an entirely different paradigm than what mainstream offerings are
View quoted post on X
june Verified account @arjunkmrm
Tardigrade v0.58.0 comes with a fully Effect native plugin system for agents called libraries:

Tardigrade library = Effect service + RPC contract + prompt blocks

Application hosts can plug different service implementations using Effect layers. Libraries also let you append static/dynamic prompt blocks into the system or user message (cache aware). For example, the latest state of the current working directory or relevant memory based on user message.

This gives strong type level guarantees, first-class dependency injection and integration with the rest of your Effect codebase. All fully durable! Currently it's closely tied to tools, but it would soon be extended to agent events. Read more here:
tardigrade.sh/docs/concepts/…
tardigrade.sh Libraries | Tardigrade Add typed capabilities and pluggable implementations to actors
5 replies 4 reposts 87 likes 2.6K views View on X
june Verified account @arjunkmrm
As you peel all the layers of what we call an agent harness, what you're left with at the end is an append-only log. This is your pet.

Compute can be ephemeral and viewed as a versioned interpreter of the log, through functions that describe what happens when the state of the log changes. This is your cattle.

Separating state and compute this way is the architecture that will scale to the gazillion agents of the future.
Marko Kraemer Verified account @markokraemer Nov 4
I think the right architectural bet for the gazillion of AI Agent sandbox providers popping up is is to separate compute from storage

cracking one shared filesystem that infinite compute writes to at local speed is needed for swarms of agents collaborating and working at scale

compute become stateless & disposable entirely. Everything that makes up the machine lives in the storage layer. The storage layer is a distributed, versioned (copy on write) filesystem whose source of truth is object storage
View quoted post on X