memento is a fast, local Rust CLI for indexing project files and personal workspace memory, then retrieving relevant context with semantic search.

It is inspired by OpenViking, with a narrower CLI-first workflow focused on local files, local memory, and a single-user developer setup.

• Index local text files into a workspace-managed vector store
• Store user and agent notes under stable mem:// URIs
• Search indexed content with local embedding models
• Inspect, list, read, reindex, and remove tracked items
• Run entirely on your machine with a local loopback server

memento init [--model <model>] [--port <port>]
memento serve
memento doctor
memento models
memento add [--force] <path>...
memento rm <path|uri>
memento remember <uri> (<text> | --file <source>)
memento forget <uri>
memento reindex [path]...
memento ls [uri]
memento cat <uri>
memento show <uri>
memento find <query>

curl -LsSf https://raw.githubusercontent.com/dagnele/memento/main/install.sh | bash

& ([scriptblock]::Create((irm "https://raw.githubusercontent.com/dagnele/memento/main/install.ps1")))

Or download a release from GitHub Releases.

Build the CLI:

cargo build

Initialize a workspace in the current directory:

cargo run -- init

Start the local server in another terminal:

cargo run -- serve

Index some project files:

cargo run -- add README.md src

Search for relevant context:

cargo run -- find "workspace config loading"

Store a user memory item:

cargo run -- remember mem://user/preferences/writing-style.md "Prefer concise technical explanations"

Inspect what is stored:

cargo run -- ls
cargo run -- show mem://user/preferences/writing-style.md
cargo run -- cat mem://user/preferences/writing-style.md

Running memento init creates a local .memento/ workspace:

.memento/
  config.toml
  index.db
  user/
  agent/

• config.toml stores workspace settings such as embedding model and server port
• index.db stores indexed items, metadata, and vectors
• user/ stores Memento-owned user memory files
• agent/ stores Memento-owned agent memory files

Tracked resources use mem://resources/... URIs. Stored notes use mem://user/... and mem://agent/....

memento uses local embedding models via fastembed.

• Default model: bge-base-en-v1.5
• Supported models:
  • bge-small-en-v1.5 - fast lightweight indexing on local machines
  • bge-base-en-v1.5 - balanced default for English notes and docs
  • bge-large-en-v1.5 - highest-quality English retrieval
  • jina-embeddings-v2-base-code - code-heavy repositories and source search
  • nomic-embed-text-v1.5 - longer English notes and general semantic search
  • bge-m3 - multilingual content across mixed repositories
• List models with cargo run -- models
• Pick one during setup with cargo run -- init --model <model>

Downloaded model files are cached locally. By default, the cache lives under ~/.memento/models, or you can override it with MEMENTO_MODEL_CACHE_DIR.

Check workspace health:

cargo run -- doctor

Reindex changed files:

cargo run -- reindex src/main.rs

Remove a tracked resource without deleting the source file:

cargo run -- rm src/main.rs

Delete a stored memory item and its backing file:

cargo run -- forget mem://agent/skills/refactor-cli

Useful commands while working on the project:

cargo fmt --check
cargo build
cargo test
cargo clippy --all-targets --all-features -- -D warnings

The project is currently an MVP-oriented local CLI. It is intentionally small in scope: local-first, single-binary, and focused on fast setup and simple command-driven workflows.