MemoryGraph
Graph-based Memory CLI for AI Coding Agents
A graph-based memory system that gives AI coding agents persistent memory.
Store patterns, track relationships, retrieve knowledge across sessions.
Using with Coding Agents
Coding agents already execute shell commands, so they can use MemoryGraph directly. Two steps: install globally, then add instructions.
Install
cd memory-graph/ts && bun install && bun link
# or: bun build src/cli.ts --compile --outfile ~/.local/bin/memorygraph
memorygraph stats # verify
memorygraph is now available from any directory. Data persists in ~/.memorygraph/.
Add Agent Instructions
Paste this into your agent's instruction file (~/.claude/CLAUDE.md, .cursorrules, .windsurfrules, or AGENTS.md):
## Memory
`memorygraph` CLI is installed. Use it for persistent memory across sessions.
Before work: `memorygraph recall --query "<task>" --limit 10` and `memorygraph briefing`
On decisions/fixes: `memorygraph store --type solution --title "<title>" --content "<what>" --tags "<component>,fix"`
On errors: `memorygraph store --type error --title "<error>" --content "<details>" --tags "<component>,error"`
Link: `memorygraph link <from-id> <to-id> SOLVES --strength 0.8`
Session end: `memorygraph store --type conversation --title "Session: <topic>" --content "<summary>" --tags "<tags>"`
Types: solution | problem | code_pattern | fix | error | workflow | command | technology
Tags: lowercase, hyphenated, include component (auth, database, cli), 2-5 per memory
Do NOT wait to be asked. Store automatically on triggers.
Example Session
# Start: load context
memorygraph recall --query "authentication redis" --limit 10
memorygraph briefing
# Fix a bug: store problem + solution, link them
memorygraph store --type problem --title "Auth token expiry too short" --content "Tokens expiring after 1h" --tags "auth,bug"
# → abc-123
memorygraph store --type solution --title "Extend token expiry to 24h" --content "Changed JWT expiry, added refresh rotation" --tags "auth,fix" --importance 0.8
# → def-456
memorygraph link def-456 abc-123 SOLVES --strength 0.9
# End: store summary
memorygraph store --type conversation --title "Session: auth token fix" --content "Fixed token expiry, added refresh rotation" --tags "auth,session-summary"
Multi-Project Setup
All projects share ~/.memorygraph/falkordblite.db by default. To isolate per project:
export MEMORY_FALKORDBLITE_PATH=~/.memorygraph/my-project.falkor
# or tag memories with project name and filter:
memorygraph search --query "auth" --tags my-project
Troubleshooting
- Agent not using commands: Use "REQUIRED"/"MUST" in instructions, include exact commands
- Command not found:
bun link from ts/, or ensure ~/.local/bin is on PATH
- Memories not persisting:
memorygraph config to check path, memorygraph health to verify
Why MemoryGraph?
Graph Relationships Make the Difference
Flat storage (CLAUDE.md, vector stores) keeps memories as isolated entries. Graph storage connects them:
[timeout_fix] --CAUSES--> [memory_leak] --SOLVED_BY--> [connection_pooling]
| |
+------------------SUPERSEDED_BY------------------------+
Query: "What happened with retry logic?" returns the full causal chain, not just individual memories.
When to Use What
CLI Commands
Memory Operations
Context Search
Intelligence
Analytics
Proactive
Integration
Temporal
Data Management
Backends
Backend Configuration
# Use FalkorDBLite (default, zero-config)
bun run src/cli.ts stats
# Use SQLite
MEMORY_BACKEND=sqlite bun run src/cli.ts stats
# Use FalkorDB (client-server)
MEMORY_BACKEND=falkordb MEMORY_FALKORDB_HOST=localhost MEMORY_FALKORDB_PORT=6379 bun run src/cli.ts stats
# Use Memgraph
MEMORY_BACKEND=memgraph MEMORY_MEMGRAPH_URI=bolt://localhost:7687 bun run src/cli.ts stats
# Use Cloud
MEMORY_BACKEND=cloud MEMORYGRAPH_API_KEY=mg_your_key bun run src/cli.ts stats
Environment Variables
Memory Types
Relationship Types
Memory Best Practices
Session Workflow
# Start of session: recall recent context
bun run src/cli.ts recall --query "recent work" --limit 10
# During work: store decisions and patterns
bun run src/cli.ts store \
--type solution \
--title "Use JWT for auth" \
--content "JWT tokens with 24h expiry, refresh token rotation" \
--tags "auth,security,api" \
--importance 0.8
# Link it to a prior decision
bun run src/cli.ts link <new-id> <prior-id> BUILDS_ON --strength 0.9
# End of session: store summary
bun run src/cli.ts store \
--type conversation \
--title "Session: auth refactor" \
--content "Refactored auth middleware, added JWT, fixed token refresh bug" \
--tags "auth,session-summary"
# Export backup
bun run src/cli.ts export --format json --output session-backup.json
Add to CLAUDE.md or AGENTS.md
## Memory Protocol
### REQUIRED: Before Starting Work
You MUST use `recall` before any task. Query by project, tech, or task type.
### REQUIRED: Automatic Storage Triggers
Store memories on ANY of:
- Git commit: what was fixed/added
- Bug fix: problem + solution
- Architecture decision: choice + rationale
- Pattern discovered: reusable approach
### Memory Fields
- Type: solution | problem | code_pattern | fix | error | workflow
- Title: Specific, searchable
- Content: Accomplishment, decisions, patterns
- Tags: project, tech, category (required)
- Importance: 0.8+ critical, 0.5-0.7 standard, 0.3-0.4 minor
- Relationships: Link related memories when they exist
Architecture
Project Structure
memory-graph/
├── ts/
│ ├── src/
│ │ ├── cli.ts # CLI entry point (35+ commands)
│ │ ├── index.ts # Library exports
│ │ ├── config.ts # Configuration management
│ │ ├── database.ts # Database interface
│ │ ├── models.ts # Data models and schemas
│ │ ├── backends/ # Backend implementations
│ │ │ ├── falkordb-shared.ts # Shared FalkorDB base class
│ │ │ ├── falkordblite.ts # Embedded FalkorDBLite
│ │ │ ├── falkordb.ts # Client-server FalkorDB
│ │ │ ├── bolt-shared.ts # Shared Bolt protocol base
│ │ │ ├── memgraph.ts # Memgraph (Bolt protocol)
│ │ │ ├── sqlite.ts # SQLite fallback
│ │ │ ├── cloud.ts # Cloud REST API
│ │ │ └── factory.ts # Backend factory
│ │ ├── tools/ # CLI tool handlers
│ │ ├── intelligence/ # Entity extraction, pattern recognition, context retrieval
│ │ ├── analytics/ # Graph visualization, similarity, learning paths
│ │ ├── proactive/ # Session briefing, predictions, outcome learning
│ │ ├── integration/ # Context capture, project analysis, workflow tracking
│ │ ├── migration/ # Backend-to-backend migration
│ │ ├── sdk/ # Cloud API client SDK
│ │ └── utils/ # Export/import, validation, helpers
│ ├── tests/ # 97 tests
│ └── package.json
├── docs/ # Documentation
└── CLAUDE.md # Agent instructions
Building
cd ts
bun install
bun test # Run tests
npx tsc --noEmit # Type check
bun build src/cli.ts --compile --outfile memorygraph # Compile binary
Import / Migration
# Import from JSON export
bun run src/cli.ts import --input backup.json --skip-duplicates
# Migrate between backends
bun run src/cli.ts migrate --to sqlite --to-path ./local.db
bun run src/cli.ts migrate --to falkordblite --to-path ./graph.falkor --no-verify
bun run src/cli.ts migrate --to memgraph --to-uri bolt://localhost:7687
SDK
The TypeScript SDK provides a client for the MemoryGraph Cloud API:
import { MemoryGraphClient } from "memorygraph/sdk";
const client = new MemoryGraphClient({ apiKey: "mg_..." });
const memory = await client.createMemory({
type: "solution",
title: "Fixed timeout issue",
content: "Used exponential backoff with retries",
tags: ["redis", "timeout"],
});
License
MIT License - see LICENSE.
Start simple. Upgrade when needed. Never lose context again.