Resilient AI Research
Resilient agents, stable control, durable memory

Resilient AI systems that recover under stress

Resilient AI Research builds open source resilience controllers (like Resilient-GNN-Controller) and memory systems that help agents retain what matters and recover from perturbations.

11×

Chaos survival improvement

62%

Success under stress

6

Resilience layers

Open

Source code

Works with Claude, GPT, and open source models.

GymnasiumPettingZooStable-Baselines3PyTorchClaudeGPTLlamavLLM

How it works

Memory that stays useful over time.

A simple request path: enrich with memory, forward to the model, then convert important outcomes into durable memory.

1

Connect any provider

Pass your provider key + endpoint. We proxy requests without locking you in.

2

Retrieve memory

Short-, mid-, and long-term memory are pulled just-in-time for the request.

3

Convert and prioritize

We compress, transform, and rank memories so the model only sees what matters.

4

Write back learnings

Important outcomes are converted into durable memory for future tasks.

Capabilities

A memory layer for any LLM app.

Short-, mid-, and long-term memory with conversion and importance filtering baked into the request pipeline.

Short-, mid-, and long-term memory

A single interface for working memory, episodic memory, and durable knowledge.

Automatic memory conversion

Summarize and reshape raw context into useful, compact memories over time.

Importance filtering

Learn what's worth keeping. Drop noise. Keep signal.

Provider-agnostic proxy API

Claude, GPT, and open source models through the same request path.

Open source-first

Transparent primitives you can self-host, audit, and extend.

Built for production apps

Designed for latency, consistency, and evolving user context.

Open source

Transparent building blocks.

We ship primitives you can self-host, audit, and extend. Swap providers without rewriting your app.

Explore projects

Resilient-GNN-Controller

A biologically-inspired resilient RL controller: coping hormones, bounded state, and recovery across adversarial environments.

memory-proxy

Provider-agnostic proxy API with memory enrichment (coming soon).

memory-sdk

Client SDKs for integrating memory into apps quickly (coming soon).

Research

Applied research, shipped as product.

We focus on making memory predictable: what gets stored, how it’s converted, and how it’s retrieved.

View research →

Resilience controllers for RL agents

Bounded-state safety and recovery

Biologically-inspired coping signals

Graph-based multi-agent control

FAQ

Questions, answered.

Clear answers about providers, memory conversion, and self-hosting.

Does this work with any LLM?+

Yes. You pass your provider key + endpoint and the proxy enriches requests with memory before forwarding.

What is "memory conversion"?+

It's the process of transforming raw interaction history into compact, reusable memory (summaries, facts, preferences) that stays useful over time.

Can I self-host?+

That's the goal. The core primitives are designed to be open and deployable in your environment.

Do you store my provider API keys?+

V1 is designed so keys can be passed per request. The website waitlist does not collect provider keys.

Join the waitlist

We’re shipping a provider-agnostic memory proxy API plus open source components for memory conversion and importance filtering.

Join the waitlist

Get early access updates for Memory Proxy, memory conversion, and SDK releases.