AI/ML●●●Banger
LLM inference slowdown fixed (177 experiments, +37% attention) – in 48h
Fused int4 attention kernel on Metal keeps LLM speed constant as context grows.
WizardrySolve My ProblemBig Brain
christinetyip
102mo ago
Back to browse
GitHub Repository
A Python library that solves context window degradation in long-running LLM agents by moving memory management out of the model layer and into a deterministic engine.
18 starsPython
Mnesis – Lossless context management Python library
by damatar·Mar 22, 2026·1 point·0 comments
AI Analysis
●●SolidBig BrainWizardry
Engine-managed memory beats model self-summarization on OOLONG benchmarks.
Strengths
- •Deterministic engine handles memory instead of unreliable model summarization
- •30-40% accuracy improvement at 8K to 1M token context lengths
- •Production-ready with CI, coverage, and OpenSSF Scorecard attestation
Weaknesses
- •Only 2 stars indicates limited real-world validation
- •Context management space has emerging competing solutions
Category
Target Audience
Developers building long-horizon LLM agent systems
Similar To
LangChain memory · LlamaIndex context managers · Mem0
Similar Projects
AI/ML●●●Banger
Tarvos – Relay Architecture for infinitely building with coding agents
40-line baton handoffs keep AI agents at peak performance across long tasks.
Big BrainZero to One
Photon48
203mo ago
Developer Tools●●Solid
Tarvos – Coding agents that work infinitely
Relay architecture beats context rot, but depends entirely on Claude Code CLI for now.
Big BrainShip It
Photon48
203mo ago
AI/ML●●Solid
Tau Router – Using Number-Theory to Partition Long-Context Retrieval
Number theory instead of learned routers, but conjecture-level math and limited production evidence.
Big BrainWizardryBold Bet
davidames0523
203mo ago
AI/ML●●Solid
Contexa – Git-inspired context management for LLM agents
Git metaphor for agent memory is clever; execution and adoption remain unproven.
Big BrainWizardry
0x0003r
103mo ago
AI/ML●●●Banger
AI agents who prevent context drift through gossip
Agents cross-review each other's work to stop context drift before it hits the wiki.
Big BrainCozyZero to One
najmuzzaman
101mo ago