@cloudhead.io relatedly www.roc-lang.org does have a very interesting memory management model although reading radiant.computer/log/002-inte... I suspect it may not be giving enough control

How is memory managed in Radiance ?

syntax and language primitives as content addressed data like unison would be amazing

Even better if syntax is just data so you can use any desired notation like isomorf did

Cloud was appealing when it was the only way to collaborate with others but when collaborating with an agent cloud seems like like an inconvenient ceremony. I don't want to go to place to collaborate with a machine, I want it meet me where I am to collaborate with me.

Personally I have being wishing not just local issue tracking system for collaborating with agents, but also a local pull request system so I have a better interface for managing work agents do. Broadly yet a project management system as a pace to manage tools/resources allocated for ongoing tasks

Interesting to observe that #local-first has a resonance with LLMs aided workflows. Need for sophisticated undo is definitely a big factor. Increasingly I'm suspecting that lack of control in feudal web creates a friction that is easier to overcome on local device under your control

Seems like 1:n vs 1:1 relation. If many people are interested in the interview with someone that someone probably won't have time to do 1:1 with each one of them

In all fairness I should also mention that github.com/git-bug/git-... project both and has similar properties

Current paradigm: predict access patterns, optimize for them. LLMs enable the opposite: open-ended exploration without knowing what questions will be asked. This fundamental incompatibility creates conditions favoring shift from service-oriented to user-oriented model
gozala.io/dialog/parad...

isomorf.io was an interesting language (which I think is dead now) where syntax was just projection of AST and you could choose between different notations.

In rust you can go pretty wild with macros

Dialog aspires to inspire exploration, so that it's fun to play with your data. Nothing flashy to share there yet, but even in these early days we are starting explore
bsky.app/profile/cdat...

Dialog synthesizing decades of computer science research

- Semantic data model (from Datomic)
- Probabilistic Search Trees
- Datalog
- Commodity Blob Stores (S3/IPFS)
- Web Assembly

It is also an embeddable database, but instead of rigid schemas you define concepts and query for facts from which those can be derived. Concepts can be remapped, adapted via datalog rules making it possible to facilitate cross-tool without their coordination

Dialog is designed around local-first principals. It is a tool like git, but for your semantic (facts, concepts) and episodic (events, observations) memory. It's goes from local to global using git like remotes represented by commodity blob store (like S3,R2,IPFS)

We have demoed dialog's ability to perform partial query-driven replication at #local-first conf in the hallway track

bsky.app/profile/cdat...