Vector Databases · verygoodplugins

automem

AutoMem is a Python-based memory service that combines graph and vector storage to give AI assistants persistent, relational memory across sessions. It uses FalkorDB for typed relationships and Qdrant for semantic search, enabling AI to recall not just matching facts but the reasoning and context behind decisions.

Talk to DEV.co Discuss Implementation

Source: GitHub — github.com/verygoodplugins/automem

771

GitHub stars

Forks

Python

Primary language

MIT

License (OSI-approved)

Key facts

Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.

Field	Value
Repository	verygoodplugins/automem
Owner	verygoodplugins
Primary language	Python
License	MIT — OSI-approved
Stars	771
Forks	98
Open issues	13
Latest release	v0.16.1 (2026-07-07)
Last updated	2026-07-07
Source	https://github.com/verygoodplugins/automem

What automem is

AutoMem implements a dual-layer storage architecture: FalkorDB maintains a canonical graph with 11 typed relationship edges (RELATES_TO, LEADS_TO, PREFERS_OVER, etc.), while Qdrant provides vector-based semantic recall. A hybrid ranking algorithm combines semantic similarity, graph traversal, temporal alignment, and importance scoring; consolidation cycles (decay, creative, cluster, forget) implement biological memory dynamics to prune weak memories and strengthen important ones.

Quickstart

Get the automem source

Clone the repository and explore it locally.

terminalbash

1git clone https://github.com/verygoodplugins/automem.git2cd automem3# follow the project's README for install & configuration

Need it deployed, integrated, or customized instead? DEV.co ships production installs.

Best use cases

Multi-Tool AI Assistant Memory

Unify memory across Claude, Cursor, ChatGPT, and other AI services. AutoMem acts as a central memory backend, allowing an assistant to recall context and reasoning from prior interactions regardless of which service is used.

Large-Context Scaling (100k+ tokens)

AutoMem degrades gracefully across large haystack sizes (57.4% at 10M tokens). Use for applications requiring long-term memory retrieval where context windows are insufficient or expensive to fill with raw data.

Solo Developers and Small Teams

Self-hosted option on Docker or Railway with no managed SaaS fees. Ideal for indie developers, researchers, and small teams who own their infrastructure and want full data sovereignty.

Implementation considerations

Requires FalkorDB and Qdrant as external dependencies; plan database provisioning, backups, and operational overhead.
Consolidation cycles (decay, creative, cluster, forget) run on configurable schedules; tune CONSOLIDATION_*_INTERVAL_SECONDS to balance memory freshness against compute cost.
Graph traversal depth and relationship expansion tunable via expand_relations, relation_limit, expansion_limit on recall endpoint; requires domain-specific tuning for optimal multi-hop discovery.
Embedding model choice (default: FastEmbed bge-base-en-v1.5) affects recall quality; changing embedders requires re-indexing all memories.
FalkorDB is the source of truth; Qdrant failure gracefully degrades to graph-only recall, but FalkorDB downtime returns 503 service error.

When to avoid it — and what to weigh

Compliance-Heavy Deployments — No built-in SOC2, HIPAA, or audit logging mentioned. Avoid if you need row-level ACLs, compliance dashboards, or enterprise SLAs. Consider Mem0, Letta, or Zep instead.
Managed SaaS Preference — AutoMem requires self-hosting (Docker, Railway, or bare metal). If you want a polished managed dashboard, hands-off operations, and vendor support, look elsewhere.
Multi-Agent Swarms with Isolation — No mention of per-agent memory isolation or multi-tenant architecture. If you need strict isolation between multiple agents sharing infrastructure, architecture may not be suitable.
Highest Conversational Recall Priority — On neutral Agent Memory Benchmark, AutoMem trails on conversational tasks (74.4% vs 94.6% on LongMemEval Core-3). If you need top verbatim recall over efficiency, choose Hindsight or alternatives.

License & commercial use

MIT License. Permissive OSI-approved license allowing commercial use, modification, and distribution with attribution. No copyleft restrictions.

MIT permits commercial use without requiring disclosure or licensing fees. However, verify that FalkorDB and Qdrant (both dependencies) also permit your intended commercial deployment. No explicit warranty or liability clauses beyond standard MIT terms; review your legal exposure for production use.

DEV.co evaluation signals

Editorial assessment — not user reviews. Directional, with an explicit confidence level.

Signal	Assessment
Maintenance	Active
Documentation	Adequate
License clarity	Clear
Deployment complexity	Moderate
DEV.co fit	Good
Assessment confidence	High

Security considerations

Self-hosted architecture shifts security responsibility to operator. No built-in encryption, authentication, or audit logging mentioned. Network transport security (TLS for FalkorDB/Qdrant connections) must be configured separately. Data isolation in multi-tenant scenarios not addressed. For sensitive data, implement perimeter authentication, encrypt credentials, and isolate database instances.

Alternatives to consider

Mem0

Managed SaaS with SOC2/HIPAA compliance, multi-agent isolation, and enterprise support. Choose if compliance and hands-off operations matter more than data sovereignty.

Letta (formerly Autogen)

Full-featured multi-agent framework with integrated memory, state management, and tool composition. Better for swarms and complex agentic workflows; steeper learning curve.

Zep

Managed memory service with RAG, long-term retention, and compliance features. Choose if you want SaaS simplicity without self-hosting infrastructure.

Software development agency

Build on automem with DEV.co software developers

AutoMem suits teams building multi-tool AI assistants, long-context applications, or self-hosted memory backends. Review the benchmark methodology, test the reproduction script, and run a proof-of-concept on Railway or Docker. Not a fit for compliance-heavy or managed SaaS-first deployments.

Talk to DEV.co

Related open-source tools

Surfaced by semantic similarity across the DEV.co open-source index.

SimpleMem LightMem ReMe

Related on DEV.co

Explore the category and the services that help you build with it.

Vector Databases AI Application Development API Development Cloud Deployment Talk to DEV.co

automem FAQ

Can AutoMem work if Qdrant is unavailable?

Yes, in degraded mode. FalkorDB serves recall via graph traversal alone. If FalkorDB is down, the service returns 503. FalkorDB is the canonical source of truth.

What embedding model does AutoMem use by default?

FastEmbed local bge-base-en-v1.5 (768-dimensional). No API keys required. Embedders are swappable but require re-indexing all memories.

How does memory consolidation work?

Four cycles: Decay (daily, exponential relevance scoring), Creative (weekly, non-obvious connection discovery), Cluster (monthly, pattern generation), and Forget (off by default, archives weak memories <0.2, deletes very old <0.05). All tunable via environment variables.

What are the benchmark trade-offs?

AutoMem excels at large-context scaling (57.4% at 10M tokens) and efficiency (~2.6–4.8k context tokens fed to answerer). On conversational recall, it trails the top board leader (74.4% vs 94.6% on LongMemEval Core-3). Pick it for scale and efficiency, not top verbatim recall.

Work with a software development agency

DEV.co is a software development agency delivering custom software development services to companies building on open source. Our software developers and web developers design, integrate, and ship production systems — spanning web development, APIs, AI, data, and cloud. If automem is part of your vector databases roadmap, our team can implement, customize, migrate, and maintain it.

Evaluate AutoMem for Your AI Memory Layer

Talk to DEV.co Schedule a Consultation