Memory for Your AI Assistants _

> Persistent Agent memory that just works!
> Your AI remembers conversations, learns patterns, grows smarter
> Simple config. Sensible defaults. Local-first. Privacy by default.

$ bunx memorygraph store --type solution --title 'Redis timeout fix' Read the Docs

Cloud Backend is live — memory across all your environments

Loading...
// automatic_memory

Memory is stored and recalled automatically as you work

claude-code ~ auth-service
you> Add rate limiting to the /api/users endpoint
claude> memorygraph recall --query "rate limiting"
→ Found: Rate limit pattern (Redis) - used in /api/auth
claude> I'll use the same Redis-based approach from /api/auth...
[implements rate limiting]
claude> memorygraph store --type pattern --title "Rate limit /api/users"
✓ Pattern stored automatically on commit
// explicit_memory

Or explicitly store and recall important solutions

claude-code ~ session
you> Remember the Redis timeout fix we just implemented
claude> memorygraph store --type solution --title "Redis timeout fix"
✓ Memory stored: mem_a1b2c3d4
you> Have we solved timeout issues before?
claude> memorygraph recall --query "timeout solutions"
✓ Found 3 memories:
→ Redis timeout fix (solution) - 0.92 relevance
→ Connection pool tuning (pattern) - 0.87 relevance

See it in action

MemoryGraph in action - Claude Code recalling priority projects and their status
Real example: Claude recalls project status and priorities across sessions

// QUICKSTART

> Get started in under a minute

Default (FalkorDBLite) RECOMMENDED

Embedded graph database with native Cypher. No server needed.

# Install via bun
bunx memorygraph store --type solution --title "Hello"

# Or clone and build
git clone https://github.com/gregorydickson/memory-graph.git
cd memory-graph/ts && bun install

# Store your first memory
bun run src/index.ts store \
  --type solution --title "Use Zod for validation"

# Configure memory directives (important!)
# Add to ~/.claude/CLAUDE.md

Configure memory directives to enable automatic storage.

With SQLite OPTIONAL

Zero-dependency fallback using Bun's built-in SQLite.

# Use SQLite backend (zero deps)
export MEMORY_BACKEND=sqlite

# Store and recall memories
bun run src/index.ts store \
  --type solution --title "Redis timeout fix"

bun run src/index.ts recall --query "timeout" --limit 5
With Cloud Backend OPTIONAL

Sync memories across devices and teams via MemoryGraph Cloud.

# Set cloud backend
export MEMORY_BACKEND=cloud
export MEMORYGRAPH_API_KEY=mg_your_key_here
export MEMORYGRAPH_API_URL=https://memorygraph.dev

# Sync memories to cloud
bun run src/index.ts store \
  --type solution --title "Auth refactor"

# Recall across all your devices
bun run src/index.ts recall --query "auth"

Pro tip: Build a standalone binary for fast CLI access: bun build src/cli.ts --compile --outfile memorygraph Learn more →

// CORE_FUNCTIONS

> CLI commands that give your AI persistent memory

memorygraph store 
memorygraph store \
  --type solution \
  --title "Redis timeout fix" \
  --tags redis,timeout

Capture solutions, patterns, decisions. Your AI builds a knowledge base from every interaction.

memorygraph recall 
memorygraph recall \
  --query "timeout solutions" \
  --limit 10

Fuzzy search with semantic matching. Find relevant context even with imprecise queries.

memorygraph link 
memorygraph link \
  --from mem_problem_123 \
  --to mem_solution_456 \
  --type SOLVES

Link problems to solutions, causes to effects. Build a knowledge graph, not just a list.

memorygraph related 
memorygraph related \
  --id mem_error_789 \
  --max-depth 2

Traverse the graph. Find what caused this, what solved that, what works with what.

$ Why Knowledge Graphs?

> Traditional vector stores lose relationships between memories

> FalkorDB-powered graphs preserve causality and context

> Ask "What solves X?" not just "What's similar to X?"

> Cloud backend auto-discovers connections via similarity detection and transitive inference

// VALIDATED_BY_RESEARCH

"Agents remember what was said but not what was learned."

— Evo-Memory, Google DeepMind (Dec 2025)
Google DeepMind's benchmark shows vector-based memory (like Mem0) provides ≤2% improvement over no memory at all. MemoryGraph's typed relationships enable true experience reuse.
[Read the Research] → [See Benchmark Results] →

// WHY_DEEP_MEMORY_MATTERS

"
Ablation studies clearly show that the depth of the memory architecture is crucial. Modules with deeper memories consistently achieve lower perplexity... and exhibit better scaling properties.
— Google Research, Titans (December 2025)
Vector databases store flat embeddings. MemoryGraph stores structured knowledge:
Shallow Memory (Competitors)
Deep Memory (MemoryGraph)
✗ Memory → [0.23, 0.45, ...]
✓ Memory → Node + Typed Edges
✗ "What solved X?" → Maybe relevant
✓ "What solved X?" → Exactly right
✗ No relationships
✓ SOLVES, DEPENDS_ON, SIMILAR_TO
✗ Time-based decay
✓ Importance-based consolidation
✗ Flat search space
✓ Graph-guided navigation
Read the Titans Paper →
// BENCHMARKED_AGAINST_COMPETITION

Google DeepMind's Evo-Memory benchmark measures how well AI memory systems enable "experience reuse" — learning from past sessions.

Environment Mem0 No Memory Delta
AlfWorld 0.27 0.28 -1%
ScienceWorld 0.32 0.31 +1%
PDDL 0.10 0.08 +2%
BabyAI 0.54 0.52 +2%

Mem0 provides ≤2% improvement over no memory at all.

Source: Evo-Memory, Google DeepMind (Dec 2025) (arXiv:2511.20857v1)

[See Full Comparison] → [Read the Research] →

// DATA_PORTABILITY

> Your data. Your choice. No lock-in.

Migrate between any backend with full verification and rollback. Start local, scale to cloud, or move between providers—your memories follow you.

🔄

Backend Migration

Move memories between SQLite, Neo4j, Memgraph, FalkorDB, and more with a single command.

✓

Verified Transfers

Automatic validation ensures every memory and relationship transfers correctly.

↩

Safe Rollback

If anything fails, automatic rollback protects your data. Always test with --dry-run first.

📦

Export Anywhere

Export to JSON anytime. Your data is never trapped in a proprietary format.

Supported Backends:
SQLiteNeo4jMemgraphFalkorDBFalkorDBLite
// quick_start
# Validate migration
$ memorygraph migrate --to falkordb --dry-run
# Execute migration
$ memorygraph migrate --to falkordb --verbose
# Verify new backend
$ memorygraph health
✓ Migration completed successfully!
Migrated: 150 memories
Migrated: 342 relationships
Duration: 3.2 seconds
Migration Guide →

// CLOUD_SYNC

Share memories across environments, devices, and teams with our managed cloud backend.

// PRICING_PLANS

> Start free. Scale as you grow. No hidden fees.

PRO RECOMMENDED
$5 /month
  • [✓] 100,000 memories
  • [✓] Advanced search
  • [✓] Graph visualization
  • [✓] API access
  • [✓] Priority support
$ ./init --pro
ULTRA
$50 /month
  • [✓] 500,000 memories
  • [✓] Advanced search
  • [✓] Graph visualization
  • [✓] API access
  • [✓] Priority support
$ ./init --ultra
TEAM
$100 /month
  • [✓] Unlimited memories
  • [✓] Up to 10 users
  • [✓] Team workspaces
  • [✓] SSO & SAML
  • [✓] Dedicated support
$ ./init --team

> Self-hosting available. Open source CLI is free forever.

// FAQ

> man memorygraph