0

corellia

niche

A formally-typed, event-sourced recursive LLM workflow engine for software creation tasks; promising but early and over-engineered for simple needs.

Source ↗
corellia social preview

What it is

A TypeScript framework implementing a recursive software factory: typed goals drive an engine that decides, splits, executes, judges, repairs, and emits reports, all persisted via an append-only event log.

How it differs from vanilla Claude

A vanilla agent can be prompted to follow a recursive plan-execute-judge-repair loop, but it lacks typed contracts, deterministic projections (memory, tree), built-in repair escalation, and an auditable event log. Corellia enforces these structurally.

Skill, plugin, or workflow shift?

High: spans contract types, recursive engine, multiple event stores (JSONL, Postgres, in-memory), LLM brain interface, check library, goal registry, pattern store, and CLI demo. Requires grasping all to leverage effectively.

Devil's advocate — is this just complexity?

A simple Python/Typescript script with a loop and an LLM call can replicate the core behavior for most tasks. The typed goal system adds ceremony without proven benefit for small teams. The event log is a nice-to-have audit trail, but not essential for development. This looks like an academic exercise, not a practical tool.

The honest case for it

For teams building a regulated, long-lived software pipeline (e.g., generating legal compliance code), the event-sourced foundation guarantees replayability and deterministic projections. The repair escalation and pattern flywheel (deferred) could make it self-improving over time. No other project offers this architecture for LLM workflows.

Who it's for

Audience fit

Primarily forAI-first Engineer
AI-first Engineer70

Depth and leverage for a technical engineer who wants to understand it and level up their workflow — not just offload work.

Vibe Coder20

Value for someone who wants a more capable tool without the technical depth — accessible, does-it-for-you.

Requires understanding of typed systems, event sourcing, and recursive architecture to use effectively. Not a 'one prompt does it all' tool.