Skip to content
vcarlPublic

About

Kareda is a research project building an agent with introspective salience: one capable local LLM acts, and a parallel "subconscious" stream reads that same model's own generation internals (token surprisal, entropy, attention, hidden-state trajectories — "j-space") to figure out what's worth remembering.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

3 Commits

Folders and files

Repository files navigation

kareda

A semi-autonomous agent with introspective salience.

One capable model acts. A parallel stream reads that model's own generation telemetry — surprisal, commitment, decision depth, groundedness — and derives salience from it. The subconscious is the exhaust of the model you already ran, not a second, weaker model.

  • docs/FOUNDATION.md — architecture and the governing thesis
  • docs/IMPLEMENTATION-PLAN.md — build order, gates, delegation
  • docs/PRECONDITIONS.md — Phase 0 falsification results (authoritative)
  • smoke/ — discriminative smoke probes and their recorded output

Status: Phase 0. Nothing is built. The thesis is not yet validated.

About

Kareda is a research project building an agent with introspective salience: one capable local LLM acts, and a parallel "subconscious" stream reads that same model's own generation internals (token surprisal, entropy, attention, hidden-state trajectories — "j-space") to figure out what's worth remembering.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages