Skip to content

INSTALLATION.md

Latest commit

 

History

History
176 lines (130 loc) · 5.09 KB

INSTALLATION.md

File metadata and controls

176 lines (130 loc) · 5.09 KB
Raw

LLMem Installation

How to install and set up LLMem. Back to README

Go Installation (recommended)

The Go binary provides the full CLI, dream cycle, and extraction.

Build from source

git clone https://github.com/MichielDean/LLMem.git
cd LLMem
make build
make install    # copies binary to ~/.local/bin/llmem

Or install the CLI manually:

go build -o ~/.local/bin/llmem ./cmd/llmem

Initialize

llmem init                  # interactive — detects providers
llmem init --non-interactive  # use defaults, no prompts

Agent integration (skills + plugins)

After building the CLI, install the agent integration layer:

cd LLMem && npm install

This runs the postinstall script which:

  1. Copies all skill directories to ~/.agents/skills/
  2. Auto-detects your agent platform (OpenCode, Claude Code, or Copilot CLI)
  3. Deploys the correct plugin to the right location

Force a specific platform:

node install.js --platform opencode      # OpenCode only
node install.js --platform claude-code   # Claude Code / Copilot CLI only
node install.js --platform both          # OpenCode + Claude Code / Copilot CLI
node install.js --platform none          # Skills only, no plugins

Platform-specific setup

OpenCode: The plugin is auto-deployed to ~/.config/opencode/plugins/llmem.js. Optionally add to your opencode.json:

{ "plugin": ["llmem"] }

Claude Code / Copilot CLI: The plugin is auto-deployed to ~/.claude/plugins/llmem/. Enable it:

claude plugin install ~/.claude/plugins/llmem
# Or test without installing:
claude --plugin-dir ~/.claude/plugins/llmem

What to install

Goal What to install
Use the llmem CLI only make build && make install
Use LLMem with OpenCode CLI + npm install (deploys skills + OpenCode plugin)
Use LLMem with Claude Code / Copilot CLI CLI + npm install (deploys skills + agent plugin)
Just the skill files (no CLI, no plugin) Copy skills/ to ~/.agents/skills/ manually

Requirements

  • Go 1.26.1+ (for building the CLI)
  • Node.js 20+ (for plugin installation only)
  • Ollama running locally for extraction, embedding, and dreaming — or set OPENAI_API_KEY / ANTHROPIC_API_KEY for cloud providers — or use FTS5-only mode without any provider

Python Installation (legacy)

The Python package is still available but the Go binary is recommended for CLI use. The Python package provides the same CLI plus a Python library for programmatic access.

From source

git clone https://github.com/MichielDean/LLMem.git
cd LLMem
pip install .

With optional extras:

# Vector similarity search (sqlite-vec)
pip install ".[vec]"

# Local embedding without any server (sentence-transformers)
pip install ".[local]"

# Both + dev dependencies
pip install ".[vec,local,dev]"

Initialize and verify:

llmem init
llmem stats

Go Package

Use LLMem as a Go library in your own projects:

go get github.com/MichielDean/LLMem
import (
    "github.com/MichielDean/LLMem/internal/store"
    "github.com/MichielDean/LLMem/internal/embed"
    "github.com/MichielDean/LLMem/internal/retriever"
    "github.com/MichielDean/LLMem/internal/metrics"
    "github.com/MichielDean/LLMem/internal/urlvalidate"
)

ms, err := store.NewMemoryStore(store.StoreConfig{
    DBPath:         "",               // defaults to ~/.config/llmem/memory.db
    VecDimensions:  0,               // defaults to 768
    DisableVec:     false,            // set true to skip vec0 virtual table
    RegisteredTypes: nil,             // defaults to 7 standard types
})
if err != nil {
    log.Fatal(err)
}
defer ms.Close()

// Embedding engine (Ollama client with LRU cache)
eng, err := embed.NewEmbeddingEngine(embed.EmbeddingConfig{})

// Hybrid search retriever (FTS5 + semantic with RRF fusion)
r, err := retriever.NewRetriever(retriever.RetrieverConfig{Store: ms, Embedder: eng})

// Embedding quality metrics
m, err := metrics.ComputeMetrics(embeddings, labels, 0)

// SSRF-protected URL validation
safe := urlvalidate.IsSafeURL(urlStr, false)

Run tests

make test
# or
go test ./...

Key differences from Python

Feature Python Go
SQLite driver Built-in sqlite3 modernc.org/sqlite (pure Go, no CGo)
Vector search sqlite-vec Python package vec0 virtual table (pure Go via modernc)
Migrations Manual numbered SQL files pressly/goose with embedded SQL files
CLI Full-featured (24 commands) Core commands (17 commands)
Embeddings/Providers Ollama, OpenAI, Anthropic, local Ollama (/api/generate and /api/embeddings)
Reranking RRF + multi-signal FTS5-only (hybrid coming)
Session hooks Full lifecycle Full lifecycle (Go implementation)
Dream cycle Full (light, deep, REM) Full (light, deep, REM)

The Go MemoryStore shares the exact same database schema as Python. You can use them interchangeably — a database created by Python is readable by Go, and vice versa.