Having studied the current core architecture

[ryslan25500-cloud]

Having studied the current core architecture, I am very impressed with how flexibly the JavaScript and Python stack has allowed us to implement complex asynchronous call logic and integration with vector databases. As a research prototype, the system has worked 100%.

However, if we look at long-term scaling, we have clearly run into the classic problem of dynamic environments. Due to the lack of strict static typing (TypeScript: 0.0%), the costs of State Drift, manual calibration of memory leaks, and protection of shell interfaces from prompt injections are starting to grow exponentially. We have to pull gigabytes of interpreted runtimes and write too much defensive validator code just so that the stack doesn't fall apart at the junction of runtime components. In its current form, this inevitably leads to excessive load on the processor cache and creates bottlenecks in the asynchronous pipeline.

In order for the project not to suffocate under the weight of its own technical debt, it seems that a radical step is needed — a complete transfer of the key orchestration core to Rust.

Switching to Rust will allow us to obtain uncompromising memory safety at the compilation stage without the overhead of a garbage collector, perfect alignment of the CPU cache and strict mathematical guarantees of reliability. This isolates system calls and transforms the architecture from a vulnerable interactive prototype into a fault-tolerant, default-protected and production-ready engine.

It is much more effective to fix the current logic as a successful MVP, learn from it and rebuild the core architecture completely on Rust, until the volume of accumulated legacy scripts completely blocked the development of the project. What do you think about the total restructuring of the core and migration to Rust?

[edoherty2017]

🍴

[Michiel-VandeVelde]

a full rewrite of the orchestration core in Rust feels like a very large step right now. Rewrites can easily stall progress unless there is a very clear scope, migration path, and measurable problem being solved, might start with certain parts tho, make an issue :)

Closing this discussion for now so larger architecture proposals can be tracked in a more focused issue/RFC.