Updated May 2026
Supermemory positions itself as "context infrastructure for your agents" — a hosted memory API, an MCP server, and connectors for sources such as Google Drive, Notion and OneDrive. Developers store and retrieve memory through the API and a search endpoint that advertises sub-300 ms p50 latency. Pricing is usage-based: roughly $0.005 per 1,000 tokens stored, $0.005 per 1,000 search queries, and separate rates for richer content and operations. The Free tier bundles $5/month of usage with the Supermemory MCP; Pro is $19/month with about $20 of usage and 2 teammates; Scale is $399/month; Enterprise is custom.
Supermemory also exposes an end-user "second brain" surface and the Supermemory MCP, so a non-developer can connect it to Claude or Cursor. Its centre of gravity, though, is the developer console and API: capabilities are shipped as endpoints you compose into your own product, and the metered model assumes you can reason about token volume. It is memory plumbing you build on, not an application you live in.
Alma is a complete persistent memory product. You sign up at alma.olivares.ai, chat with Claude (Haiku, Sonnet or Opus 4.7 with 1M-token context), and the memory layer captures facts, preferences and decisions automatically. The Soul Engine handles identity. Image, video, music, calendar and documents are built in. Developers can also reach the same memory through the MCP server, the JavaScript SDK and the REST API on any paid plan.
Memory is structured into three layers (memories, episodes, procedures), scored with five factors (relevance, importance, confidence, recency, frequency), and assembled into the system prompt in under 100 ms. Chat, image, voice, video and music all draw from a single monthly budget — larger on each plan, from Starter ($14/mo) to Pro ($29/mo) to Max ($99/mo) — that resets on your subscription anniversary. Alma runs on EU-edge infrastructure with the timezone set to Europe by default.
Supermemory is infrastructure: you build the agent or app, your users never see Supermemory's name, and you pay for the tokens and queries your product generates. Alma is a product: you sign up, chat, and the memory works immediately; developers are an additional audience served by the SDK, MCP server and REST API on every paid plan. If your goal is "ship my own agent with a memory backend", Supermemory is purpose-built. If your goal is "use AI with memory across the tools I already work in", Alma covers it without writing integration code.
This is the most honest point of difference, and neither model is universally cheaper. Supermemory meters usage: you pay per 1,000 tokens stored and per 1,000 queries, starting from a $5/month free allowance. That is predictable for a developer who can model token volume and wants to pay only for what an app consumes — and genuinely cheap at low volume. Alma charges a flat monthly fee with a single AI budget inside it (Starter $14/mo, Pro $29/mo, Max $99/mo), so an individual never has to reason about per-token costs. Pick metered infrastructure if you are building and can forecast usage; pick a flat product budget if you are an end user who wants one bill.
Supermemory stores memory and exposes search and SuperRAG endpoints; the composition of what lands in your model's prompt is your responsibility. Alma's 3-layer architecture is opinionated about shape: facts go in memories, conversation summaries in episodes, learned workflows in procedures, each with its own retrieval rules. Context assembly builds the final prompt for you — Soul blocks first, then memories, episodes and procedures, all within the model's token budget. You get an assembled prompt, not a list of search results to wire together.
| Feature | Supermemory | Alma |
|---|---|---|
| Primary shape | API + MCP server (context infrastructure) | Full product (web app + chat) with SDK/API on every paid plan |
| Target user | Developers building agents and apps | End users + developers |
| Web app / chat | Second-brain surface; centre of gravity is the API | Full chat with streaming, tools, file attachments |
| AI identity | Not provided — bring your own | Soul Engine (13 versioned identity blocks) |
| Context assembly | Search + SuperRAG endpoints — you compose the prompt | Built-in assembled prompt, <100 ms, 5-factor scoring |
| MCP server | Yes — Supermemory MCP | Yes — Claude Desktop / Cursor / Windsurf |
| Search latency | Sub-300 ms p50 (advertised) | <100 ms context assembly |
| Creative tools | None | Image / video / music studios in Pro and Max |
| Hosting | Not specified as EU-resident | EU-edge infrastructure, Europe timezone default |
| Pricing model | Usage-metered: Free ($5/mo) · Pro $19/mo · Scale $399/mo | Flat: Starter $14 · Pro $29 · Max $99 per month |
Building an agent with a memory backend. An engineering team wants memory their app calls over an API, billed by usage, with connectors to Drive and Notion. Supermemory is purpose-built: call the API, pay for the tokens and queries you use. Alma's SDK (on any paid plan) also covers this, but adds a full cognitive layer (Soul Engine, scored context assembly, typed memory) instead of raw search.
Personal memory across the tools you use. A solo user wants memory that follows them between Claude Desktop, Cursor and a web chat. Supermemory's MCP can do part of this but expects you to assemble the experience; Alma is the experience — sign up, chat, connect the MCP server in five minutes, and the same memory is everywhere.
Choose Supermemory if you are building an agent or product and want memory as usage-priced infrastructure. You want to call an API, control prompt composition yourself, connect sources at the API level, and pay only for the tokens and queries you consume — which is cheap at low volume. You are comfortable writing integration code and want sub-300 ms search you invoke directly. For developers who can forecast usage, the metered model is a genuine advantage over a flat fee.
Choose Alma if you want to use AI with memory rather than build the plumbing. Sign up, start chatting, and the memory layer works without code. You get chat with Opus 1M context, creative studios, calendar and documents in one place, a single predictable monthly budget instead of per-token math, and EU-edge hosting. Developers who also want an embeddable layer get the SDK, REST API and MCP server on every paid plan — with a complete cognitive layer rather than a raw memory store.
Can I use both? Yes. They are not mutually exclusive — a developer could use Supermemory inside a product and use Alma as their personal AI workspace. They target different jobs.
Does Alma have an API like Supermemory? Yes, on any paid plan: a REST API, JavaScript SDK and MCP server cover memory CRUD, hybrid search, context assembly and conversation streaming. The difference is that Alma returns an assembled cognitive layer, while Supermemory returns search results you compose.
Which is cheaper? It depends on how you use it. Supermemory's usage-metered model is cheaper at low volume and for apps that can forecast token use; Alma's flat monthly budget is more predictable for individuals who do not want to track per-token costs.
Supermemory is the right tool when you are building an agent and want memory as usage-priced infrastructure you call over an API. Alma is the right tool when you want to use AI with memory across the tools you already work in, on a flat monthly budget, without writing integration code. Starter ($14/mo) is enough to test whether Alma fits; if you also need an embeddable layer, Starter ($14/mo) already includes the SDK, REST API and MCP server with a full cognitive layer.