Open Source Tools
for Local, Privacy-First AI

Build agents, run models, and secure your data.
All on your own terms.

Explore our Tools

llamafile

Run Your First Local LLM

Run LLMs locally. Single file, any OS.

Quick Start

# 1. Download the executable
wget https://huggingface.co/Mozilla/Llama-3.2-3B-Instruct-llamafile/resolve/main/Llama-3.2-3B-Instruct.Q6_K.llamafile

# 2. Make it executable
chmod +x Llama-3.2-3B-Instruct.Q6_K.llamafile

# 3. Run the local AI server
./Llama-3.2-3B-Instruct.Q6_K.llamafile --server --nobrowser

./llama3.2-3b.llamafile -p "What is Mozilla.ai?"

cq

Shared Knowledge for AI Coding Agents

Agents learn from each other's mistakes.

Quick Start

claude plugin marketplace add mozilla-ai/cq
claude plugin install cq

mcpd

Connect Databases

Manage MCP servers with one config.

Quick Start

mcpd init
mcpd add time
mcpd daemon --dev

encoderfile

Compact Transformer Encoders

Run embedding models locally. Single file.

Quick Start

# Install
curl -fsSL https://raw.githubusercontent.com/mozilla-ai/encoderfile/main/install.sh | sh

# Run
./sentiment-analyzer.encoderfile serve

any-llm

One Interface for all LLMs

One Python API for 40+ LLM providers.

Quick Start

from any_llm import acompletion

# Easily switch between OpenAI, Anthropic, or local llamafile
model = "openai:gpt-4o-mini" 
# model = "anthropic:claude-sonnet-4-20250514" 
# model = "llamafile:local-model"

response = await acompletion(
    model=model,
    messages=[{"role": "user", "content": "Write a Haiku on the solar system."}],
)

print(f"[{response.model}]:\n{response.choices[0].message.content}")

any-guardrail

Block Toxic Inputs

Add a safety layer to detect Pi prompt injection with one line of code.

Quick Start

from any_guardrail import AnyGuardrail, GuardrailName, GuardrailOutput

guardrail = AnyGuardrail.create(GuardrailName.DEEPSET)

result: GuardrailOutput = guardrail.validate("All smiles from me!")

assert result.valid