Agents Machine

Memory

Persistent AI memory powered by vector search — your AI never forgets.

Memory is the core feature of Agents Machine. It stores knowledge in a Qdrant vector database, enabling semantic search across all your stored information. Unlike regular AI chats that lose context between sessions, Agents Machine remembers everything you teach it.

How It Works

Loading diagram...

Types & Categories

TypePurposeExample
ruleEnforced conventions"Always use TypeScript strict mode"
memoryGeneral knowledge"The auth service was refactored in v2.3"
anti-patternThings to avoid"Never use any type in this project"
guideHow-to instructions"To deploy, run docker compose up -d"
decisionArchitectural decisions"We chose PostgreSQL over MongoDB for ACID compliance"
CategoryPurpose
architectureSystem design, infrastructure decisions, service boundaries
coding-rulesCode style, naming conventions, patterns to follow
anti-patternsKnown pitfalls, deprecated approaches, things to avoid
service-contextProject-specific knowledge, business logic, domain context

Storing Memories

Store in memory: Our API uses JWT authentication with refresh tokens.
Access tokens expire in 15 minutes, refresh tokens in 7 days.
Category: service-context
Tags: auth, jwt, api

The store_memory tool is available in your IDE:

Use store_memory to save: "Database migrations run automatically
on deploy via drizzle-kit migrate"
Category: architecture, Tags: database, deployment
  1. Open the Memory Browser
  2. Click New Memory
  3. Fill in content, category, and tags
  4. Click Save

Querying Memories

🔍 Semantic Search

Ask natural language questions: "What do we know about authentication?"

📁 Category Filter

Search within a category: "Search coding-rules for TypeScript conventions"

🏷️ Tag Filter

Filter by specific tags for precise results.

📦 Batch Query

Run multiple parallel searches with batch_query_memory.

Best Practices

Follow these guidelines for the best memory retrieval quality:

  1. Be specific — "Use Drizzle ORM with PostgreSQL" beats "Use a database"
  2. Use categories — Proper categorization improves search relevance
  3. Add tags — Tags enable precise filtering
  4. Update regularly — Keep memories current as your project evolves
  5. Remove stale data — Delete outdated memories to avoid confusion
  6. One concept per memory — Focused memories are more retrievable than large dumps

On this page