Alma is a cognitive memory and identity engine for AI conversations. Instead of starting every chat from scratch, Alma automatically extracts facts (Memories), conversation summaries (Episodes), and learned behavioral patterns (Procedures) from your conversations. A 12-block Soul Engine defines your AI's identity — worldview, communication style, anti-patterns, and behavioral boundaries — and assembles personalized context for every conversation in milliseconds.
Three memory types: semantic facts scored by importance and confidence, conversation episode summaries tagged with topics and outcomes, and behavioral procedures learned automatically. Hybrid search combines keyword matching with semantic embeddings for precise recall.
Available as web app, MCP server (21 tools, 9 resources for Claude Desktop/Cursor/Windsurf), VSCode extension, and full REST API (200+ endpoints). Powered by Anthropic Claude (3 models). 15 languages. BYOK support. AES-256-GCM encryption. TOTP 2FA. OAuth (Google + GitHub). Full GDPR data export and deletion.
Free tier: 500 memories, 50 episodes, Claude Haiku — no credit card, no time limit. Pro $19/mo. Advanced $49/mo. Ultimate $149/mo.
Key Features
Persistent Memory (memories, Episodes, Procedures)
Soul Engine (12+ Identity Blocks)
Automatic Knowledge Extraction From Conversations
MCP Server (21 Tools, 9 Resources)
Full Rest API (200+ Endpoints)
Vscode Extension
Voice Input (stt) And Text-to-speech (tts)
AI Image Generation (flux Pro + Leonardo AI)
Document Generation (docx, xlsx, pptx, md, txt, html, svg)
BYOK Support (anthropic, Replicate, Leonardo API Keys)
Hybrid Search (keyword + Semantic Embeddings)
9 AI Agent Tools (web Search, Deep Research, Canvas, Documents)
15-language Localization
AES-256-GCM Encryption + TOTP 2FA + OAuth
Full Data Export/import (.alma Format)
Conversation Folders And Tags
Context Environments (work/personal Separation)
Give your AI a soul
Alma is a persistent memory layer for AI. Your conversations build knowledge over time — preferences, facts, decisions, patterns. Every interaction makes your AI smarter, more personal, more yours.
01 — Process
How it works
1
Choose your path
Use Alma the way that fits you best.
Web App
Recommended
Sign up and start chatting instantly. No downloads, no setup.
Extensions
VSCode and browser to integrate Alma into your workflow.
API / MCP
For developers. Full REST API and MCP server.
2
Converse
Chat naturally. Alma extracts memories, facts, and preferences automatically.
3
Evolve
Every conversation enriches your AI's understanding. It grows with you.
02 — Features
Everything your AI needs to remember
A complete memory system — not just a vector store. Alma understands context, personality, and learned behavior.
Persistent Memory
Your AI remembers facts, preferences, decisions, and patterns across every conversation.
Context Assembly
Automatically assembles the right context for each interaction. No manual prompt engineering.
Soul Engine v2
Define identity, worldview, style guide, anti-patterns, and communication modes. Not just personality — a complete cognitive framework.
Environments
Separate memories by context — work, personal, projects. Each with its own soul.
Image Studio & Voice
Dedicated Image Studio with AI art generation (Flux Pro, Leonardo AI). Voice chat with speech-to-text and text-to-speech. File uploads with AI vision analysis.
Developer-First
Full REST API, VSCode extension, and MCP server out of the box.
Export Anything
Export conversations to PDF, DOCX, HTML. Memories to XLSX. Full GDPR data dumps.
Smart Search
Keyword, semantic, or hybrid search across all your memories with date and category filters.
Bring Your Own Keys
Use your own API keys for Anthropic, Replicate, and Leonardo AI. Full control over costs and model selection.
03 — Integrations
Works with your favorite tools
Web App
Chat with memory — No setup
MCP Server
Claude Desktop & Cursor
VSCode
IDE extension
Image Studio
Flux Pro & Leonardo AI
JavaScript SDK
Node.js & TypeScript
REST API
140+ endpoints
04 — Pricing
Simple, transparent pricing
Start free. Upgrade when you need more memory and power.
Product Overview
The Complete Memory Layer
for AI
Alma is a persistent memory system that transforms stateless AI into a genuine collaborator. It captures facts, summarizes conversations, and learns behavioral patterns — building a cognitive profile that grows with every interaction. Your AI stops forgetting and starts understanding.
3-Layer Memory Architecture
Memories
facts, preferences, decisions
Episodes
summaries, context, threads
Procedures
patterns, triggers, workflows.
01 — Architecture
Three layers of understanding
Alma's memory system mirrors how humans process and retain information — through facts, experiences, and learned behaviors.
1
Memories
Discrete facts about the user — preferences, decisions, technical context, personal details. Each memory is semantically indexed and retrievable by relevance.
User prefers TypeScript over JavaScript
Project uses PostgreSQL 16
Prefers concise explanations
2
Episodes
Compressed summaries of conversations and interactions. Episodes capture the arc of a discussion — what was decided, what changed, what matters.
Debugged auth middleware issue
Designed new database schema
Reviewed API architecture decisions
3
Procedures
Learned behavioral patterns and workflows. Procedures define when and how the AI should act — from communication style to complex multi-step processes.
When reviewing code, check error handling first
Use bullet points for technical explanations
Always suggest tests for new functions
Soul Engine & Context Assembly
All three layers feed into the Soul Engine, which assembles the optimal context for every interaction. It selects the most relevant memories, recent episodes, and applicable procedures — then renders them into a structured system prompt that gives the AI deep understanding of who you are and what you need.
02 — Capabilities
Everything you need for persistent AI
A complete platform for building AI that remembers, learns, and adapts to every user.
Persistent Memory
Every fact, preference, and decision is stored and recalled across sessions. No more repeating yourself. The AI remembers what matters to you.
Context Assembly
Automatically builds the right context for every conversation. Relevant memories, recent episodes, and learned procedures — assembled in milliseconds.
Soul Engine v2
A complete cognitive framework that defines how the AI thinks, responds, and grows. Personality, guidelines, and behavioral patterns — all configurable.
Multi-Model Support
3 intelligence tiers powered exclusively by Anthropic's Claude models. Free plans include Normal (Claude Haiku). Paid plans add Advanced (Claude Sonnet) and Complex (Claude Opus). Switch freely without losing context or memory continuity.
Multi-Modal
Dedicated Image Studio with Flux Pro and Leonardo AI. Voice chat with speech-to-text and text-to-speech. File uploads with AI vision analysis. Document generation (DOCX, XLSX, PPTX). All with persistent memory.
Background Processing
Learns while you chat. Memories are extracted, episodes are summarized, and procedures are refined — all happening seamlessly in the background.
03 — Technical
How it works
Four steps from zero to a fully memory-aware AI. Integrate in minutes, not weeks.
1
Connect
Generate an API key in your dashboard settings.
# Generate an API key in Settings → API Keys
# Then authenticate with Bearer token
curl https://alma.olivares.ai/api/v1/me \
-H "Authorization: Bearer YOUR_API_KEY"
2
Assemble Context
Build the perfect context for any conversation with a single API call.
# Assemble relevant context for any query
curl -X POST https://alma.olivares.ai/api/v1/context/assemble \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{ "query": "Help me refactor the auth module" }'
# Returns: memories, episodes, procedures
# ranked by semantic relevance
3
Chat with Memory
Send messages through Alma and get responses enriched with full memory context.
# Send a message with full memory context
curl -X POST https://alma.olivares.ai/api/v1/chat \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{ "message": "What approach should we take?",
"model": "claude-sonnet" }'
# The AI knows your preferences, your project
# context, and your history.
4
Background Learning
After each conversation, Alma automatically extracts and stores new knowledge.
# This happens automatically after each chat:
# 1. New facts are extracted as memories
# → "User decided to use JWT for auth"
# 2. The conversation is summarized as an episode
# → "Discussed auth refactoring approach"
# 3. Behavioral patterns are updated
# → "When discussing architecture, provide
# trade-offs and alternatives"
04 — Platform
Integrate everywhere
Alma meets you where you work. REST API, IDE plugins, MCP server — one memory layer, every surface.
REST API
Full CRUD access to memories, episodes, context, and chat
MCP Server
Model Context Protocol for Claude Desktop and IDE tools
VSCode Extension
Persistent coding context directly in your editor
JavaScript SDK
Type-safe Node.js & TypeScript client
05 — Models
Your choice of intelligence
3 intelligence tiers powered exclusively by Anthropic's Claude models. Free plans get 1 tier (Normal / Claude Haiku); paid plans get all 3 (Normal, Advanced, Complex). Switch freely without losing memory or context.
Tier Free Pro+
Normal Claude Haiku 4.5 (Anthropic) Claude Haiku 4.5 (Anthropic)
Advanced — Claude Sonnet 4.6 (Anthropic)
Complex — Claude Opus 4.6 (Anthropic)
Start building with persistent memory
Give your AI the ability to remember, learn, and grow. Free to start, scales with you.