Open Source · MIT License

Agent-Agnostic Memory
Platform for AI Agents

Memory as infrastructure. Layered architecture with reasoning-aware retrieval, built-in forgetting & compression. Make AI agents cheaper, more consistent, and genuinely improving over time.

GitHub Repository Explore Architecture

4

Memory Layers

95%

Cost Reduction

MIT

Open Source

Core Architecture

Four Layers of Memory

Inspired by human cognition. Each layer serves a distinct purpose, working together to give AI agents comprehensive memory capabilities.

Working Memory

Short-term · Context-aware

Temporary storage for immediate context and current task information. Like a human's mental scratchpad — holds what the agent needs right now and automatically clears when tasks complete.

Episodic Memory

Event-based · Temporal

Stores experiences and events with temporal context. Enables agents to recall past interactions, learn from outcomes, and build a rich history of what happened and why.

Semantic Memory

Knowledge · Concepts

The agent's encyclopedia. Stores factual knowledge, domain concepts, and general information. Knowledge that doesn't depend on when or where it was learned.

Procedural Memory

Skills · Procedures

Stores learned procedures, skills, and how-to knowledge. Enables agents to perform complex multi-step tasks reliably — muscle memory for AI.

Key Innovations

What Makes It Different

Not just another vector database. ThinkingMemory is purpose-built memory infrastructure for AI agents.

01

Reasoning-Aware Retrieval

Agents don't just search for keywords — they ask what they need to remember. ThinkingMemory understands the reasoning context and returns the most relevant memories, not just the most similar vectors.

Context-aware memory retrieval
Intent-based query understanding
Cross-layer memory fusion

// Agent asks what it needs to remember
memory.recall({
    context: "user asking about pipeline failure",
    intent: "diagnose root cause",
    layers: ["episodic", "semantic"]
})

// Returns contextually relevant memories
// → Past failure patterns (episodic)
// → Pipeline topology knowledge (semantic)
// → Resolution procedures (procedural)
                

02

Built-in Forgetting & Compression

Just like human memory, not everything needs to be remembered forever. ThinkingMemory automatically compresses old memories, forgets irrelevant details, and consolidates patterns — keeping storage costs minimal and retrieval fast.

Intelligent memory decay algorithms
Pattern consolidation over time
Automatic storage optimization

// Configure memory lifecycle
memory.configure({
    working: { ttl: "session" },
    episodic: {
        compress_after: "7d",
        consolidate_after: "30d",
        forget_threshold: 0.1
    },
    semantic: { deduplicate: true },
    procedural: { versioned: true }
})
                

03

Agent-Agnostic APIs

Works with any AI framework and any LLM. REST APIs and SDKs that plug into LangChain, CrewAI, AutoGen, custom agents, or anything else. Your memory infrastructure shouldn't lock you into a framework.

REST API + Python/JS SDKs
Framework-agnostic integration
Works with any LLM provider

// Works with any framework
from thinkingmemory import Memory

mem = Memory(agent_id="my-agent")

// Store a memory
mem.store(
    layer="episodic",
    content="Pipeline X failed at 3am",
    metadata={"severity": "high"}
)

// Retrieve with reasoning context
memories = mem.recall(
    "Why did the pipeline fail?"
)
                

Why ThinkingMemory

Transform Your AI Agents

95% Cost Reduction

Agents without memory repeat computations, re-fetch context, and waste tokens. Memory eliminates redundant processing, dramatically cutting API costs.

Consistent Behavior

No more agents that forget previous interactions or give conflicting responses. Persistent memory ensures reliable, context-aware behavior across all sessions.

Continuous Improvement

Agents learn from past experiences, improving performance over time without retraining. Every interaction makes the agent smarter and more capable.

Framework Independence

Swap frameworks without losing memory. Switch LLM providers without rebuilding. Your agent's knowledge persists regardless of the stack underneath.

Get Started Today

Give Your Agents Memory

Join the open-source revolution in AI memory infrastructure. Make your agents cheaper, more consistent, and genuinely improving over time.

Cloud Option Available: Don't want to self-host? We offer a fully managed cloud version at nominal costs. Contact us for pricing.

GitHub Repository Contact Sales

MIT License Framework Agnostic Production Ready

Agent-Agnostic Memory Platform for AI Agents