{ "name": "lgrep", "version": "0.2.1", "description": "Semantic code search CLI with local and Postgres-backed cloud profiles", "type": "module", "main": "dist/index.js", "module": "dist/index.js", "types": "dist/index.d.ts", "exports": { ".": { "types": "./dist/index.d.ts", "import": "./dist/index.js" } }, "bin": { "lgrep": "dist/cli/index.js", "lgrep-mcp": "dist/mcp/index.js" }, "files": [ "dist//*.js", "dist//.js.map", "dist/**/.d.ts", "dist/viewer/**", "README.md", "LICENSE" ], "scripts": { "start": "bash ./scripts/start-hosted-server.sh", "build": "tsup && npm run build:viewer", "build:test-binaries": "node scripts/ensure-test-binaries.mjs --force", "dev": "tsup --watch", "build:viewer": "vite build --config viewer/vite.config.ts", "dev:viewer": "vite --config viewer/vite.config.ts", "pretest": "node scripts/ensure-test-binaries.mjs", "test": "vitest run", "test:cloud-smoke": "bash scripts/cloud-smoke.sh", "test:watch": "vitest", "test:coverage": "vitest run --coverage", "lint": "eslint src --ext .ts", "type-check": "tsc --noEmit", "release:check": "npm run type-check && npm run build && npm test && npm pack --dry-run", "prepublishOnly": "npm run type-check && npm run build && npm test" }, "engines": { "node": ">=18.17" }, "keywords": [ "search", "semantic-search", "code-search", "embeddings", "ollama", "anthropic", "openai", "groq", "ai", "llm", "local", "cli", "vector-search", "rag", "code-intelligence", "developer-tools" ], "author": "Dennison Bertram dennison@dennisonbertram.com", "license": "MIT", "repository": { "type": "git", "url": "git+https://github.com/dennisonbertram/lgrep.git" }, "homepage": "https://github.com/dennisonbertram/lgrep#readme", "bugs": { "url": "https://github.com/dennisonbertram/lgrep/issues" }, "publishConfig": { "access": "public" }, "dependencies": { "@ai-sdk/anthropic": "^2.0.56", "@ai-sdk/cohere": "^2.0.21", "@ai-sdk/groq": "^2.0.33", "@ai-sdk/openai": "^2.0.87", "@babel/parser": "^7.28.5", "@babel/traverse": "^7.28.5", "@babel/types": "^7.28.5", "@lancedb/lancedb": "^0.13.0", "@modelcontextprotocol/sdk": "^1.25.1", "@solidity-parser/parser": "^0.20.2", "ai": "^5.0.113", "apache-arrow": "^21.1.0", "chalk": "^5.3.0", "chokidar": "^5.0.0", "commander": "^12.1.0", "ignore": "^7.0.5", "ollama": "^0.5.11", "open": "^11.0.0", "ora": "^8.1.1", "pg": "^8.13.1", "tree-sitter": "^0.25.0", "tree-sitter-c": "^0.24.1", "tree-sitter-cpp": "^0.23.4", "tree-sitter-go": "^0.25.0", "tree-sitter-java": "^0.23.5", "tree-sitter-python": "^0.25.0", "tree-sitter-rust": "^0.24.0", "voyageai": "^0.1.0", "zod": "^4.2.1" }, "devDependencies": { "@eslint/js": "^9.39.2", "@radix-ui/react-select": "^2.2.6", "@radix-ui/react-separator": "^1.1.8", "@radix-ui/react-slot": "^1.2.4", "@radix-ui/react-switch": "^1.2.6", "@tailwindcss/postcss": "^4.1.18", "@types/babel__traverse": "^7.28.0", "@types/node": "^22.10.2", "@types/pg": "^8.11.10", "@types/react": "^19.2.7", "@types/react-dom": "^19.2.3", "@typescript-eslint/eslint-plugin": "^8.50.0", "@typescript-eslint/parser": "^8.50.0", "@vitejs/plugin-react": "^5.1.2", "@vitest/coverage-v8": "^2.1.8", "autoprefixer": "^10.4.23", "class-variance-authority": "^0.7.1", "clsx": "^2.1.1", "cytoscape": "^3.33.1", "cytoscape-fcose": "^2.2.0", "eslint": "^9.39.2", "lucide-react": "^0.562.0", "postcss": "^8.5.6", "react": "^19.2.3", "react-dom": "^19.2.3", "s3rver": "^3.7.1", "tailwind-merge": "^3.4.0", "tailwindcss": "^4.1.18", "tsup": "^8.3.5", "typescript": "^5.7.2", "vite": "^7.3.0", "vitest": "^2.1.8" } }

lgrep

AI-powered semantic code search. Find code by meaning, not just text.

Search for "authentication logic" and find OAuth handlers, JWT validation, and session management — even if those words never appear in the code. Built-in code intelligence finds dead code, circular dependencies, and shows blast radius before refactoring.

Install

Requires Node.js >= 18.17.

npm install -g lgrep
lgrep init

lgrep init walks you through the two supported first-run paths:

• Local - local index and local cache
• Cloud - Postgres-backed index and Postgres-backed cache

From Source

git clone https://github.com/dennisonbertram/lgrep && cd lgrep
npm install --legacy-peer-deps
npm run build
node dist/cli/index.js init

Quick Start

lgrep init
lgrep doctor
lgrep list
lgrep search "user authentication logic"
lgrep search --usages "validateUser"
lgrep search --definition "UserService"
lgrep context "add rate limiting"
lgrep intent "what calls awardBadge"

Agent Integration

lgrep install --target claude
lgrep install --target codex
lgrep install --target mcp
lgrep install --target all
lgrep install --target all --global --server-url https://lgrep.example.com --server-auth-token <token>

Targets:

• claude installs the Claude skill, the SessionStart hook, and repo-local CLAUDE.md guidance by default
• codex writes guidance into AGENTS.md or ~/.codex/AGENTS.md when --global is used
• mcp configures lgrep as an MCP server
• all installs all three

Session start behavior:

• In local mode, the Claude SessionStart hook can clean stale state and start local watchers automatically.
• In cloud mode, the SessionStart hook verifies the hosted service is reachable and tells agents when a repo is simply unbound on first run, with concrete next steps (lgrep project list, lgrep worktree bind, or local indexing).
• Codex currently uses installed AGENTS.md instructions as its startup ritual. OpenAI’s documented Codex surfaces today are AGENTS.md and notify, not a SessionStart-style hook.
• For Codex or manual sessions, start with lgrep first. Use lgrep doctor for local setup/debugging, or lgrep worktree resolve plus lgrep project list for hosted connectivity and first-run binding checks.

For a machine-wide hosted setup, use:

lgrep install --target all --global \
  --server-url https://lgrep.example.com \
  --server-auth-token <token>

That persists the hosted URL and token into the active lgrep profile, updates ~/.claude/CLAUDE.md, updates ~/.codex/AGENTS.md, and carries the same hosted settings into the MCP config.

The intended workflow is:

1. Install lgrep integration.
2. Let the SessionStart hook prepare local or hosted lgrep automatically when Claude starts.
3. If hosted, use lgrep worktree resolve to confirm the current worktree binding. If it says no match, treat that as a first-run unbound repo, not a generic outage.
4. Use lgrep search, lgrep callers, lgrep impact, or lgrep context before broad rg/grep searches.

Commands

Core

Code Intelligence

Analysis & Exploration

All commands support --json for scripting. Most support -i, --index and -l, --limit.

Embedding Providers

lgrep config model auto                 # auto-detect (default)
lgrep config model voyage:voyage-code-3 # explicit

LLM Providers (Summarization)

Auto-detected. Priority: Groq > Anthropic > OpenAI > Ollama.

lgrep config summarizationModel auto                   # default
lgrep config summarizationModel groq:llama-3.1-8b-instant # explicit

Project Config

Create .lgrep.json in your repo root to skip --index flags:

{
  "index": "my-project",
  "root": "src"
}

Remote Storage

For cloud mode, lgrep defaults to Postgres for both the index and the cache. S3/R2 is still supported as an advanced/manual path, but it is no longer the default onboarding route.

See docs/guides/remote-storage.md for setup.

Hosted Query Service Preview

You can also run lgrep behind a shared HTTP query service instead of giving every agent direct database credentials.

Fastest Self-Hosted Path

For a Mac mini, Hetzner box, or any SSH-accessible machine:

export LGREP_DATABASE_URL="postgres://user:password@host:5432/lgrep"

lgrep server install-remote user@host \
  --server-url https://lgrep.example.com

That will:

• install lgrep on the remote host
• install a supported Node runtime when the host's system node is too old
• install Linux build/runtime prerequisites when the host needs them
• provision a launchd, working systemd, or tmux service runner
• create a remote token-store file for hosted auth
• configure this machine globally for Claude, Codex, and MCP unless you pass --skip-local-install

For the exact Hetzner/Mac-mini runbook, including the tmux tunnel/bootstrap flow for many local worktrees, see docs/guides/self-hosted-ssh-runbook.md.

Project Bootstrap

Then register a repo and its worktrees in the same hosted Postgres:

export LGREP_DATABASE_URL="postgres://user:password@host:5432/lgrep"

lgrep server bootstrap /path/to/repo \
  --project repo-main \
  --branch main \
  --worktree 'feature-login|/path/to/repo-feature-login|feature/login'

Client Usage

After a global install, this machine can just use lgrep directly:

lgrep project info repo-main
lgrep worktree list --project repo-main
lgrep search "authentication flow" --project repo-main
lgrep callers createSession --project repo-main --worktree feature-login

If you want to configure another client machine manually:

export LGREP_SERVER_URL="https://lgrep.example.com"
export LGREP_SERVER_AUTH_TOKEN="your-hosted-service-token"
lgrep project info repo-main
lgrep worktree list --project repo-main
lgrep search "authentication flow" --project repo-main
lgrep callers createSession --project repo-main --worktree feature-login
lgrep impact createSession --project repo-main --worktree feature-login
lgrep context "trace session token flow" --project repo-main --worktree feature-login

Hosted semantic search and hosted context need a real embedding provider on the server, such as OPENAI_API_KEY. Hosted search --definition and search --usages remain database-backed client flows today.

If you want MCP clients to use the hosted service without shell exports, run:

lgrep install --target mcp --global \
  --server-url https://lgrep.example.com \
  --server-auth-token <token>

Today this hosted path is a bearer-token-protected, single-tenant query layer for project/worktree discovery, semantic search, callers, impact, and context packages. For self-hosted SSH deployments, lgrep server install-remote provisions a filesystem-backed token store on the remote host, which is the recommended auth path today. For Railway and other stateless remote deployments, keep using the service-wide LGREP_SERVER_AUTH_TOKEN. See docs/guides/hosted-query-service.md for the hosted multi-worktree workflow, SSH self-host path, Railway deploy path, and current limits.

If You Want To Use Hosted Mode Today

This is the practical next step if your goal is "one hosted database, many worktrees":

1. Create a cloud profile that points at Postgres.
2. Create one project for your repo.
3. Create one worktree per branch or checkout you want searchable.
4. Run lgrep server start on the machine that has Postgres access.
5. For self-hosted deployments, you can mint a scoped token with lgrep server token create.
6. For Railway and similar remote deployments today, use the shared LGREP_SERVER_AUTH_TOKEN service secret.
7. Point agents at LGREP_SERVER_URL plus LGREP_SERVER_AUTH_TOKEN.

That is enough to start using hosted lgrep now.

What We Are Building Next

The next product step is not another setup step for you. It is feature expansion on top of the hosted query service:

• hosted HTTP MCP transport
• richer hosted code-intel coverage like definitions/usages and deeper impact analysis
• background sync and hosted indexing workers

Those will make the hosted path more useful for agents, but they are not required for the current multi-worktree setup.

Programmatic API

import { createEmbeddingClient, createAIProvider, detectBestProvider } from 'lgrep';

const embedder = createEmbeddingClient({ model: 'auto' });
const { embeddings } = await embedder.embed(['hello world']);

const ai = createAIProvider({ model: detectBestProvider() });
const explanation = await ai.generateText('Explain this code...');

Configuration

By default, lgrep reads config from the active profile. You can manage profiles with lgrep profile.

lgrep profile list
lgrep profile create cloud
lgrep profile use cloud
lgrep config                   # show all settings
lgrep config model             # get one
lgrep config model auto        # set one
lgrep doctor                   # check everything

You can still override everything with LGREP_HOME for a one-off isolated home:

LGREP_HOME="$HOME/Library/Application Support/lgrep-local" lgrep doctor
LGREP_HOME="$HOME/Library/Application Support/lgrep-local" lgrep index . --name my-project

License

MIT

Contributing

git clone https://github.com/dennisonbertram/lgrep && cd lgrep
npm install --legacy-peer-deps
npm run build
npm test

Maintainers: see docs/guides/releasing.md for npm release setup and the tag-based publish flow.