wagl

This repository contains the wagl CLI: a local-first agent memory store.

What wagl is

• Local-first memory store backed by libSQL
• Designed for automation/agents: all commands emit JSON
• Includes curation + “expertise areas” for maintaining shareable, structured knowledge

Installation

cargo install --path crates/cli --locked --force

Core commands (memory)

# init schema
wagl init

# insert
wagl put --type note --text "hello" --tag demo --salience 0.7

# get
wagl get <id>

# query
wagl query "hello" --limit 10

# status
wagl status

# audit/import file-based memories
wagl audit --root /path/to/workspace
wagl import-missing --root /path/to/workspace

# search across primary + all expertise areas
wagl everything "term" --limit 50 --include-index

Global options

• --db <PATH>: override the wagl DB file path (also supported via env WAGL_DB).

Integrations

Hive/event consumption and notification delivery have moved out of this repo (they live in the worker). This repo is focused on the local-first memory CLI.

Curation (shareable subset DB)

Wagl can produce a curated DB (subset) from a source DB using a TOML config.

# create a shareable subset DB
wagl curate --config docs/CURATE_CONFIG.example.toml --out /tmp/wagl-curated.db --overwrite

• Example config: docs/CURATE_CONFIG.example.toml
• Use-case: generate a smaller DB safe to expose to other tools (e.g., MCP) or a specialized agent.

Expertise areas

Expertise areas are “secondary collections” stored in the same wagl DB.

# list expertise areas
wagl expertise list

# add an expertise area
wagl expertise add cms-quality-measures --description "…" --when-to-use "…"

# add an item to an expertise area
wagl expertise put --expertise cms-quality-measures --type note --text "…" --tag source:doc --salience 0.5

# query an expertise area
wagl expertise query cms-quality-measures dqm --limit 20

Repo layout

• crates/core — types + validation (no IO)
• crates/db — libSQL schema + queries
• crates/cli — clap CLI entrypoints

Vector embeddings (optional)

wagl supports optional semantic search via sqlite-vec, statically linked at build time — no external .so file required. To enable, set your embeddings provider:

export WAGL_EMBEDDINGS_BASE_URL="http://localhost:11434"
export WAGL_EMBEDDINGS_MODEL="nomic-embed-text"

If neither WAGL_EMBEDDINGS_BASE_URL nor WAGL_EMBEDDINGS_MODEL is set, wagl runs in text-only mode — no errors, no startup cost. See docs/sqlite-vec.md for details.

Acknowledgements

• sqlite-vec by Alex Garcia — MIT/Apache-2.0 licensed. Statically linked for vector similarity search.

Help

Run:

wagl --help