TencentDB-Agent-Memory: Fully Local Long-Term Memory for AI Agents

Overview

TencentDB-Agent-Memory (8,094 stars) is an open-source solution that delivers fully local long-term memory for AI agents using a 4-tier progressive pipeline — with zero external API dependencies. Developed by Tencent Cloud, it enables agents to remember past interactions, learn from experience, and maintain context across sessions, all while keeping data entirely on-device.

The project has rapidly grown to 8K+ stars and 581 stars gained today, reflecting the community's urgent need for practical, privacy-preserving agent memory solutions.

Features

• 4-tier progressive pipeline: Stages memory from working → short-term → long-term → archival, optimizing for both speed and retention
• Fully local operation: No external API calls, no cloud dependencies, no data leaving the device
• Zero LLM dependency: Memory management runs independently without requiring external language models
• Persistent storage: Memory survives agent restarts, session boundaries, and system reboots
• Semantic retrieval: Search past agent experiences by meaning, not just keywords, using local embeddings

How It Works

The 4-tier pipeline processes agent experiences progressively:

Tier	Retention	Detail	Access Speed
Working	Session-only	Full detail	Instant
Short-term	Hours	Summarized	Sub-millisecond
Long-term	Days-weeks	Consolidated with semantic index	Milliseconds
Archival	Indefinite	Compressed	Hundreds of ms

Installation

pip install tencentdb-agent-memory

Quick Start

from tencentdb_agent_memory import AgentMemory

# Initialize local memory store
memory = AgentMemory(storage_path="./agent_memory")

# Store an agent experience
memory.store(
    experience="Fixed authentication bug in login flow",
    context={"repository": "auth-service", "severity": "high"},
    agent_id="coding-agent-1"
)

# Retrieve relevant past experiences
results = memory.query(
    "How did we fix auth issues before?",
    agent_id="coding-agent-1",
    limit=5
)

# Memory persists across sessions
for result in results:
    print(f"Past experience: {result.experience}")

Why It Matters

Agent memory remains one of the hardest unsolved problems in production AI systems. Most current approaches either:

1. Rely on cloud APIs — introducing latency, cost, and privacy concerns
2. Use simple context windows — losing information between sessions
3. Require complex infrastructure — making deployment difficult

TencentDB-Agent-Memory's fully local approach addresses all three issues simultaneously, making it an attractive option for privacy-sensitive applications, offline deployments, and cost-constrained environments.

Pros

• ✅ Fully local — no privacy concerns or API costs
• ✅ Progressive pipeline optimizes storage and retrieval efficiency
• ✅ Zero external LLM dependencies for memory management
• ✅ Active development by Tencent Cloud team

Cons

• ❌ Relatively new project with evolving API
• ❌ Best suited for Python-based agent frameworks
• ❌ May need tuning for specific agent architectures

When to Use

Use TencentDB-Agent-Memory when:

• Your agents need persistent memory across sessions
• Privacy requirements prevent cloud-based memory solutions
• You're deploying agents in offline or air-gapped environments
• You want to minimize operational costs for agent memory

Resources

• GitHub Repository