PYLIT // SELF-IMPROVING SYSTEMS
Intelligent systems that help humans and machines think together
PYLIT

Accelerating scientific discovery with self-improving systems.

Pylit builds self-improving systems that help humans and AI teams work together on longer, harder, more consequential problems by compounding execution evidence into reusable logic.

pylit show platforms
Platforms // focus returns here
Keys: ↑/↓ choose, → focus Open Platform, ← return to list, Enter open new tab, ? toggle help, Esc close. Focus is held in the terminal while this screen is open.
COMPOUND KNOWLEDGE

OpenDream

Persistent memory for research systems: traces, reviews, and graph-backed context that survive across sessions.

Purpose Memory substrate for long-running AI research.
Signal Traces, reviews, and graph-backed learning.
Best for Teams that need persistent context.
Open platform ↗
System architecture / compounding loop v0.3
A five-stage scientific discovery loop in which goals generate candidate artifacts, evaluation tests them, refinement improves the best candidates, validated knowledge compounds, and promoted artifacts raise the next starting point. GOALS + HYPOTHESES questions, constraints, priors 01 GENERATE ideas / code / experiments / proofs 02 EVALUATE tests / simulations / benchmarks / verifiers 03 SELECT + REFINE keep best / fix weak / recombine / iterate 04 COMPOUND KNOWLEDGE memory / skills / prompts / tools / datasets 05 PROMOTE only validated artifacts become reusable BETTER SOLUTIONS NEW DISCOVERIES
Research systems in active development
01 / Why science

Humanity advances when intelligence compounds.

Science is civilization’s most important compounding machine. It turns observation into theory, theory into experiment, experiment into evidence, and evidence into new capability. Self-improving systems should help compress that loop without discarding rigor, skepticism, or human judgment.

02 / The loop

Progress compounds when execution evidence updates reusable artifacts.

The useful abstraction is not that an AI improves itself. It is that a system converts evidence into better candidates, memories, skills, tools, harnesses, programs, and verification contracts.

Generatehypotheses, proofs, programs, plans
Executeexperiments, tools, environments
Verifytests, traces, constraints, feedback
Improverepair, mutate, recombine, abstract
Promotemake the lesson reusable
03 / What compounds

Memory, skills, tools, harnesses, and programs become the substrate of discovery.

Pylit starts with focused systems, tests them in real workflows, and keeps the parts that improve future work. The open-source work is the proving ground. Products may follow where reliability, governance, collaboration, and scale become necessary.

04 / North star

AI should expand the frontier of what humans can discover and compress not only knowledge, but time.

PYLIT // EARLY ACCESS

Help shape systems for scientific discovery.

Tell us where self-improving research systems could have the greatest practical impact. We will share relevant research, prototypes, and early access opportunities.