AI Agent
Alma
Alma is a local-first MCP server that gives AI agents a private, consent-based memory of your personal facts and preferences.
Alma
What is Alma?
Alma is an open-source, local-first MCP (Model Context Protocol) server that stores a persistent self-model for AI agents, including your name, role, working style, and preferences. It keeps data on your machine, exposes only approved parts, and makes all changes auditable.
Alma vs Similar AI Tools
| Pricing Model | Free | Free | Free, Free Trial | Free, Paid |
| Free Credits | ||||
| Key Features |
|
|
|
|
| Pros |
|
|
|
|
| Cons |
|
|
|
|
| Best For |
|
|
|
|
How to use Alma?
- 1Build from source with Cargo: 'cargo build --release'
- 2Run guided setup: './target/release/alma quickstart'
- 3Connect to a host: './target/release/alma connect cursor --apply --global' or './target/release/alma connect claude --apply'
- 4Agents interact via MCP using tools like 'alma_get_reading' and 'alma_propose_facet'
Alma Key Features
- Local-first, data stays on your machine
- Consent-based reads and writes with human approval tokens
- Append-only event log for auditability and reversibility
- Portable export as JWS-signed bundles
- Companion TUI for browsing and managing facets
- MCP protocol over stdio with scoped readings
Alma Use Cases
- Give AI agents consistent memory across different chats and tools
- Store personal preferences like name, role, and working style
- Enable privacy-respecting agent personalization
- Port self-model between vendors without vendor lock-in
Alma Pricing & Free Credits
Alma currently operates on a Free model.
This tool is completely free to use
Alma Pros & Cons
Pros
- Strong privacy: data stays local and consent-based
- Open source and auditable
- Vendor-independent portable self-model
- Granular access control for agents
Cons
- Experimental project, APIs may change
- Requires Rust toolchain to build
- No packaged releases yet
- Technical setup needed
What is Alma best for?
- Developers building privacy-conscious AI agents
- Users who want persistent agent memory without vendor lock-in
- Anyone needing local-first personal context for AI tools