Updated May 2026
Claude Memory is Anthropic's built-in personalisation feature, rolled out broadly during 2026. On Claude.ai you get project-scoped memory: each Project keeps its own context (uploaded files, custom instructions and a memory store the model maintains). Free users get a small number of Projects; Pro ($20/mo) and Max ($100/$200/mo) increase the limits. There is also a Styles feature for tone and a Project instructions block for behaviour rules.
It is a meaningful upgrade over ChatGPT Memory — project-scoping prevents work and personal contexts from contaminating each other, and the import-from-ChatGPT path is smooth. The boundary, however, is the Claude product itself. Memory does not leave Claude.ai. There is no MCP integration, no API to fetch your memories, no way to bring the same context to Cursor when you are coding or to a VSCode session. If your AI workflow lives across multiple tools, Claude Memory only covers the time you spend in Claude.ai.
Alma is a persistent memory layer for AI. Like Claude Memory, it uses Anthropic Claude — Haiku, Sonnet and Opus 4.7 (1M-token context) — as the underlying model. Unlike Claude Memory, the memory itself is independent: it lives in your Alma account and reaches every AI client you use through the MCP server (@olivaresai/alma-mcp), the VSCode extension, the JavaScript SDK and the REST API.
Memory is structured into three layers (memories, episodes, procedures), the Soul Engine provides 13 versioned identity blocks, and Environments isolate per-project contexts the same way Claude Projects do — but cross-tool. Paid plans start at $14/mo (Starter) with the full Soul Engine; Pro and Max raise the memory headroom.
The architectural difference is flat vs three-layer. Claude Memory keeps a per-Project memory store the model maintains in plain text. Alma separates knowledge into three retrieval-distinct shapes: memories (discrete facts), episodes (compressed conversation summaries) and procedures (learned workflows). Each layer has its own scoring and lifecycle, which gives the model better signal at retrieval time.
The portability difference is more obvious. Claude Memory is intrinsically tied to Claude.ai — there is no MCP integration, no API endpoint to read or write memories programmatically, no way to mirror that memory into Cursor or Windsurf. Alma's memory reaches every MCP-compatible client (Claude Desktop, Cursor, Windsurf, Claude Code) plus VSCode, plus the SDK, plus the REST API. Same memory, every tool.
The control difference: Claude Memory shows you a digest of what it knows but does not expose hybrid search, scoring breakdowns or per-memory category metadata. Alma exposes full CRUD on every memory, hybrid (semantic + keyword) search with score breakdown, importance / confidence editing, bulk operations, and full export in JSON / PDF / DOCX / XLSX or .alma archive.
Claude offers Styles (tone presets) and per-Project Instructions. Both work well for casual personalisation. Alma's Soul Engine is more structured: 13 blocks across identity, worldview, personality, expertise, communication style, anti-patterns, rules and context, organised into XML sections, versioned and always injected with priority. Each block is independently editable — you can swap personality without touching expertise, or carry the same identity into a new Environment with fresh memories.
| Feature | Claude Memory | Alma |
|---|---|---|
| Memory scope | Per-Project (limited Projects on Free) | Per-Environment (unlimited on Pro and Max) |
| Memory architecture | Flat per-Project store | 3-layer (memories + episodes + procedures) |
| Cross-platform | Claude.ai only — no MCP, no API | Claude Desktop / Cursor / Windsurf via MCP, VSCode, SDK, REST API |
| LLM models | Anthropic Claude only | Anthropic Claude Haiku / Sonnet / Opus 4.7 (1M) |
| AI identity | Styles + per-Project Instructions | Soul Engine — 13 versioned blocks across identity, personality, expertise, rules |
| Memory search | Conversation search | Hybrid (semantic + keyword) with multi-factor scoring |
| Import sources | ChatGPT exports | ChatGPT, Claude, Gemini, CSV, .alma archives |
| Export formats | Conversation transcripts | JSON, PDF, DOCX, XLSX, .alma archive (full GDPR) |
| Creative tools | Built-in Claude features | Image Studio, Music Studio, Video Studio bundled in Pro and Max |
| BYOK | N/A | Every plan, incl. Free — Anthropic, Replicate, Leonardo, ElevenLabs |
| Paid plan from | $20/mo (Claude Pro) | $14/mo (Starter), $29/mo (Pro), $99/mo (Max) |
Claude.ai Free includes basic memory; Pro is $20/mo and Max plans go to $100 and $200/mo. Memory ships bundled — you cannot subscribe to Memory alone or use it outside Claude. Alma's pricing is flatter and decoupled from any specific model: Starter $14/mo (Haiku + Sonnet, unlimited memories), Pro $29/mo (adds Opus 1M, unlimited environments, creative studios), Max $99/mo (largest monthly budget — API, MCP server, VSCode extension and SDK come with every paid plan; BYOK works on every plan, including Free). Annual billing is "pay 10 months, get 12" on every paid plan.
Coding across Claude Desktop, Cursor and VSCode. A developer uses Claude Desktop for design discussions, Cursor for implementation, and VSCode for review. Claude Memory only covers the Claude Desktop conversations. Alma + MCP gives every one of those tools the same memory account — naming patterns, architecture decisions and prior PR feedback follow the developer between tools instead of staying trapped per app.
Per-project context. Claude Projects work well for organising work inside Claude.ai — separate Project per client, separate memory store per Project. Alma Environments are the same idea but cross-tool: switch the Alma environment in your sidebar (or via the MCP server) and every connected client picks up the new context (memories, episodes, procedures, Soul). Same separation, broader reach.
Creative work alongside chat. Generating images, video and music inside Claude.ai means using the products and prices Anthropic ships there. Pro and Max plans on Alma bundle Image Studio (Flux 1.1 Pro, Leonardo), Video Studio (Runway Gen-4) and Music Studio (ElevenLabs, up to 10 min) on the same monthly USD budget that powers chat — no per-feature credit accounting and no separate creative subscription.
Claude Memory is the right call if your AI workflow lives entirely inside Claude.ai — the chat there, with the same Projects each day, no Cursor, no Windsurf, no VSCode integration. The native experience is well-built, the per-Project scoping is clean, and you do not have to set up anything. If you also want Anthropic's creative or document features inside the same product, Claude's native bundle is fine.
Choose Alma if your AI workflow extends beyond Claude.ai. If you code in Cursor, write in Claude Desktop, ship via Claude Code or use Windsurf, you need a memory layer that follows you into all those tools — that is what the Alma MCP server does. Choose Alma if you want a structured, versioned AI identity (Soul Engine), if you want full export and edit control over what the AI knows, if you want bundled creative studios (Image / Video / Music), or if you need the REST API for your own integrations. The Starter plan ($14/mo) is the entry point — its monthly budget covers months of real usage with no per-feature caps.
Alma's Import workflow accepts Claude conversation exports directly: download your Claude data archive, drop it into Settings → Import, and Alma extracts memories from your Project conversations using the same extraction pipeline that powers live chat. Imports run with SSE progress, a preview-before-commit step and semantic dedup against your existing memories. Per-Project context lands in matching Alma Environments.
Does Alma replace Claude.ai? No. Alma is the memory layer; the Alma web app is one of several places you can use that memory. Claude Desktop and the Claude.ai web interface stay exactly as they are — Alma adds memory to them via MCP, it does not replace them.
Can I use Claude Memory and Alma together? Yes. Claude Memory continues to work inside Claude.ai; Alma covers everything else (Cursor, VSCode, Windsurf, Claude Code, the Alma web app). Many users start with both and migrate fully once they trust the import.
Does Alma use my data to train models? No. Alma never trains on user content. Conversations and memories are private to your account and used only to assemble context for your own future chats.
Why use Alma if Claude already supports MCP? Because the memory you connect via MCP needs a backend. Alma is that backend: structured 3-layer memory, Soul Engine identity, scoring, search, lifecycle management. MCP is the wire format; Alma is the memory product on the other end of the wire.
Claude Memory is well-built and the right call if your AI workflow lives entirely inside Claude.ai. Alma is the right call when you also work in Cursor, Windsurf, Claude Code or VSCode — same memory, every tool, via MCP. They are not mutually exclusive; many users keep both during a transition. The Starter plan ($14/mo) is enough to evaluate seriously and migrating is one click.