GitHub Repository

Crash-resistant cognitive continuity for AI agents — checkpoint/restore, cross-model state reconstruction, semantic recall, and compression. Your agent survives crashes, restarts, and model switches.

117 starsPython

AgentKeeper – Cross-model memory for AI agents

Name: AgentKeeper – Cross-model memory for AI agents
Availability: InStock
Author: thinklanceai

by thinklanceai·Feb 24, 2026·1 point·0 comments

Visit Project View on HN

AI Analysis

●●SolidSolve My ProblemBig Brain

Recovers 95% critical facts when switching GPT-4 ↔ Claude with real benchmarks.

Strengths

•Real problem: agent memory loss during crashes and provider switches is a genuine pain point.
•Concrete benchmark: 19/20 critical facts recovered bidirectionally with documented methodology.
•Clean API and SQLite persistence that actually ships—not vaporware.

Weaknesses

•Narrow scope: solves one problem (context reconstruction) that LangChain memory chains partially already handle.
•No evidence of scale testing; 100-fact benchmark is small relative to real agent workloads.

Post Description

I built AgentKeeper to solve a problem I kept hitting: every time I switched LLM providers or an agent crashed, it lost all context.

AgentKeeper introduces a Cognitive Reconstruction Engine (CRE) that stores agent memory independently of any provider and reconstructs optimal context when switching models.

Benchmark: 19/20 critical facts recovered when switching GPT-4 → Claude (and reverse). Bidirectional, tested on real API calls.

Supports OpenAI, Anthropic, Gemini, Ollama. SQLite persistence. MIT license.