🦙 LlamaCode

An AI‑powered coding assistant that runs locally via Ollama, built with TypeScript. Now includes Retrieval‑Augmented Generation (RAG) support using a SQLite‑backed vector store and a sub‑agent system for delegating complex tasks.

---

Requirements

• Node.js v18+
• pnpm
• Ollama

---

Install Ollama

macOS

brew install ollama

Or download the app directly from ollama.com/download.

Linux

curl -fsSL https://ollama.com/install.sh | sh

Windows

Download the installer from ollama.com/download.

---

Pull a Model

LlamaCode requires a model that supports tool/function calling. Recommended options:

ollama pull gpt-oss:120b-cloud
ollama pull nomic-embed-text

To see all models you have installed, run ollama list.

Once pulled, make sure Ollama is running:

ollama serve

By default, Ollama runs on http://localhost:11434. You can override this with the OLLAMA_HOST environment variable.

---

Setup

Clone the repo and install dependencies:

git clone https://github.com/fortezhuo/llamacode.git
cd llamacode
pnpm install

Set the model in src/lib/constant.ts to match what you pulled:

export const MODEL = "gpt-oss:120b-cloud"
export const EMBED_MODEL = "nomic-embed-text"

---

Usage

pnpm start

LlamaCode will start an interactive CLI session in your current working directory.

 🦙 LlamaCode

 ❯ buka package.json
 🦙 Invoking tool: read_file
 🦙 Here’s your package.json: ...

 ❯ tambahkan script build ke package.json
 🦙 Invoking tool: write_file
 🦙 Done! Added "build": "tsc" to scripts.

---

Commands

Command	Description
/help	Show available commands
/messages	Show current conversation history
/reset	Clear conversation and start fresh
/quit	Exit LlamaCode

---

Available Tools

LlamaCode gives the AI access to the following operations within your current working directory:

Tool	Description
read_file	Read the contents of a file
write_file	Write or overwrite a file
delete_file	Delete a file
bash	Execute a Bash command and capture its output
grep	Search for a pattern in files (regex)
glob	Find files matching a glob pattern
read_pdf	Extract text from a PDF file (used for RAG)
search_pdf	Perform a semantic search over indexed PDFs
sub_agent	Delegate complex or multi‑step operations to a secondary agent, enabling richer workflows and parallel tool usage

Security: All file operations are sandboxed to the directory where LlamaCode was launched. The AI cannot access files outside of it.

---

Sub‑Agent Feature

The newly added sub‑agent (src/tools/sub_agent.ts) allows the primary LlamaCode agent to spin up a lightweight secondary agent for handling long‑running or multi‑step tasks. This improves responsiveness and keeps the main conversation focused while the sub‑agent works in the background. Use the sub_agent tool when you need the AI to:

• Perform a series of dependent operations without blocking the main thread.
• Run intensive computation or data processing tasks.
• Interact with external services or APIs that require multiple steps.

The sub‑agent shares the same sandboxed environment and tool set, but operates under its own conversational context, returning results back to the primary agent when complete.

---

Configuration

Environment Variable	Default	Description
OLLAMA_HOST	http://localhost:11434	Ollama server URL

Example:

OLLAMA_HOST=http://192.168.1.10:11434 pnpm start

---

Project Structure

src/
  lib/
    agent.ts       # Main CLI loop, input routing
    chunk.ts       # Splits long text into overlapping chunks for RAG embedding
    command.ts     # Built‑in command handlers (/help, /reset, etc.)
    constant.ts    # Model, ANSI colors, tool definitions
    db.ts           # SQLite wrapper for vector store persistence
    embed.ts        # Handles text embedding using the configured model
    invoke.ts       # AI invocation and tool execution loop
    ollama.ts      # Ollama client and schema conversion
    print.ts        # Colored terminal output
    readline.ts    # User input interface
    security.ts    # Path sandboxing (safePath)
    store.ts       # Conversation history and system prompt
  tools/
    read_file.ts
    write_file.ts
    delete_file.ts
    bash.ts
    grep.ts
    glob.ts
    read_pdf.ts
    search_pdf.ts
    index.ts
    sub_agent.ts
  index.ts         # Entry point
  type.ts          # Shared TypeScript types

---

License

ISC