# mozilla.ai

> Mozilla.ai - Building trustworthy, transparent, and controllable AI

## Overview

Mozilla.ai is building a future where AI works for you. They believe the next generation of AI solutions should be trustworthy, transparent, and controllable. Mozilla.ai is backed by the Mozilla Foundation and focuses on creating infrastructure and products to build intelligent systems that align with user goals.

## Mission & Values

**Mission:** Help people work better with AI, safely, transparently, and on their own terms by building products and advancing open-source projects that make AI more trustworthy.

**Core Values:**
- **Trust by Design:** Observability, safety, and human oversight built in from the start
- **Open Foundations:** Contributing to and building on open-source for transparency and extensibility
- **Learning Together:** Testing, listening, and improving in the open
- **Agency:** AI should empower people and teams, not replace them
- **Focus:** Cutting complexity and making deliberate choices

## Products

### Octonous

An AI agent platform that automates workflows across your apps, turning scattered tools and repetitive tasks into secure, automated workflows. Teams can focus on impact and deliverables while staying in control.

**Features:**
- Connects to apps like Slack, Notion, GitHub, Confluence, Google Drive, Jira, and more
- Secure integration with internal systems
- Full visibility into agent performance with built-in evaluation and feedback
- Built on transparent, open-source foundation

**Use Cases:**
- Post Mortem Generator Agent (Support)
- Contract Prep Agent (Sales)
- Talk to Roadmap Agent (Product Management)

**Status:** Beta rollout with selected partner teams

### any-llm Platform

A simplified management and analytics platform for all LLM providers. Built on top of the any-llm library, it provides unified monitoring and management for both cloud-based and local models.

**Key Benefits:**
- **Client-side Encryption:** API keys encrypted with client-side generated key pair
- **Local and Remote Flexibility:** Track usage of both cloud providers and local models (e.g., ollama)
- **Single Key Simplicity:** One API key configured to access multiple LLM providers
- **Privacy by Default:** Only tracks metadata (token counts, model names, timestamps); no sensitive text data stored
- **Cost Tracking:** Real-time spending across providers in unified dashboard
- **Full Transparency:** Usage insights and token tracking across all LLM calls

**Status:** Currently in Beta and free to use

**Coming Soon:**
- Budget alerts for cost control
- Hosted guardrails for LLM safety
- Error and latency tracking
- Smart routing (opt-in)
- Model recommendations (opt-in)

## Open-Source Tools

### Choice-first Stack

A unified open-source stack that simplifies building and testing modern AI agents and apps through interoperable, composable tools.

### any-agent

**Description:** Universal interface for building and evaluating across agent frameworks

**Features:**
- One interface for multiple frameworks (standardized API)
- Standardized tracing using GenAI OpenTelemetry
- Trace-first evaluation (LLM-as-a-judge and Agent-as-a-judge techniques)
- Support for MCP and A2A (Agent2Agent) protocols
- Switch frameworks with configuration changes, no code rewrites

**Supported Frameworks:** Multiple leading agent frameworks

**Example:** Location-Based Assistant Demo integrating retrieval, LLM calls, and tool-based planning

### any-llm

**Description:** Library providing a single interface to various LLM providers

**Features:**
- Unified API across multiple providers (AWS, Azure, OpenAI, Anthropic, Mistral, etc.)
- Built with official SDKs for compatibility, performance, and stability
- No extra infrastructure needed (no proxies or gateways)
- Switch models and providers on demand
- Used in Mozilla.ai Agent Platform

**Why:** Solves the fragmented LLM ecosystem with unique APIs, parameters, response formats across providers

### any-guardrail

**Description:** Common interface for guardrail models to keep AI agents safe

**Features:**
- Unified interface across different guardrail models
- Quick switching between models by changing guardrail name
- Supports Prompt Injection Detection, Content Moderation, Customizable Fine-Tuned Judges
- Formalizing research code into production-ready implementations
- Seamless integration into any-agent callbacks

**Why:** Alleviates difficulty in using and evaluating guardrail models which each have their own prompts, labeling taxonomy, or fine-tuned models

### mcpd

**Description:** Toolchain and runtime layer for managing MCP servers across environments

**Features:**
- **Declarative Tool Management:** Define MCP servers, tools, and versions in .mcpd.toml config
- **Language-Agnostic:** Call Python (uvx) or JS/TS (npx) servers through unified HTTP REST API
- **Separation of Config and Secrets:** Templated secrets file kept separate from project config
- **Unified Dev Experience:** Single CLI binary for server lifecycle, requests, logs, proxying
- **Seamless SDK Integration:** mcpd_sdk in Python for native-like function calls
- **Local to Prod Without Changes:** Same config works in dev, CI, or production
- **Middleware & Observability Hooks:** Proxy layer for security, monitoring, or logic injection

**Why:** Solves the fragility and lack of scalability in manually managing modular tool servers across environments

**Installation:** Available via Homebrew (`brew tap mozilla-ai/tap && brew install mcpd`)

### llamafile

**Description:** Single-file executable for running open-source LLMs locally

**Features:**
- Bundle full LLM (model weights + inference engine + runtime) into one file
- Cross-platform (Windows, macOS, Linux, BSD) - no installation required
- Local-first and privacy-conscious (runs entirely on device, fully offline)
- Low barrier for adoption

**Use Cases:** Key enabler of sovereign AI - run models privately, offline, with full control over compute, data, and cost

**Project Type:** Mozilla Builders project

### encoderfile

**Description:** Package transformer-based encoders into a single executable binary

**Features:**
- No Python, dependencies, or runtime overhead
- Supports embedding/feature extraction, sequence classification, token classification
- Compact size (tens-to-hundreds of megabytes vs gigabytes)
- Deterministic, offline, and safe for compliance-sensitive environments
- Integration-ready as CLI, gRPC/HTTP microservice, or MCP server

**Use Cases:** Ideal for constrained or compliance-sensitive environments requiring dependency-free inference

### Lumigator (LLM Eval)

**Description:** Tool for evaluating, comparing, and selecting the best language model for projects

**Features:**
- Guided evaluation process
- Currently supports summarization and translation tasks
- Transparent insights into model performance, bias, and reliability
- Refine and optimize models for specific use cases
- Community-driven development

**Why:** Choosing the right language model shouldn't be a guessing game

### Mozilla.ai Blueprints

**Description:** Community-led collection of reusable automation blueprints

**Features:**
- Integrates open-source tools and models
- Created for developers, learners, and AI tinkerers
- Real-world use cases

## Technology Stack Integration

The Choice-first Stack components work together seamlessly:
1. Choose agent framework using **any-agent**
2. Route LLM calls via **any-llm**
3. Define tool servers declaratively with **mcpd**
4. Guard outputs and reasoning with **any-guardrail**
5. Iterate or swap components without rewriting orchestration logic

## Team & Organization

- Distributed team across 6+ countries
- 10 nationalities represented
- 12+ languages spoken
- Team includes engineers, product thinkers, researchers, and open-source builders
- Rooted in Mozilla's values
- Fast-moving SaaS company working toward product-market fit

**Leadership & Governance:** Board includes leaders from Mozilla and respected figures in AI, safety, governance, and open infrastructure

## Contact & Resources

- **Blog:** https://blog.mozilla.ai/
- **Parent Organization:** Mozilla Foundation (https://www.mozilla.org/)
- **GitHub:** Projects available on GitHub for community contribution

## Community & Contributions

Mozilla.ai actively encourages:
- Open-source contributions to their tools
- Community feedback on product development
- Collaborative development approach
- Transparency in building and testing

## Philosophy

"We're building a future where AI works for you" - Mozilla.ai believes AI should be:
- Trustworthy
- Transparent
- Controllable
- Aligned with user goals
- Empowering (not replacing) people and teams

## Links

### Core
- https://www.mozilla.ai/
- https://www.mozilla.ai/company/about-us
- https://blog.mozilla.ai/

### Products
- Octonous: https://www.mozilla.ai/product/octonous
- Agent Showroom: https://www.mozilla.ai/product/agent-platform/agent-showroom
- Octonous early access signup: https://www.mozilla.ai/sign-up
- any-llm platform (product page): https://www.mozilla.ai/product/any-llm
- any-llm platform (app): https://any-llm.ai/
- any-llm updates signup: https://www.mozilla.ai/sign-up-any-llm

### Open Tools
- Choice-first Stack: https://www.mozilla.ai/open-tools/choice-first-stack
- any-agent: https://www.mozilla.ai/open-tools/choice-first-stack/any-agent
- any-llm (library): https://www.mozilla.ai/open-tools/choice-first-stack/any-llm
- any-guardrail: https://www.mozilla.ai/open-tools/choice-first-stack/any-guardrail
- mcpd: https://www.mozilla.ai/open-tools/choice-first-stack/mcpd
- llamafile: https://www.mozilla.ai/open-tools/llamafile
- encoderfile: https://www.mozilla.ai/open-tools/encoderfile
- Lumigator: https://www.mozilla.ai/open-tools/lumigator

### Blueprints
- Blueprints home: https://blueprints.mozilla.ai/
- What are Blueprints: https://blueprints.mozilla.ai/what-are-blueprints

### Company
- Careers: https://www.mozilla.ai/company/careers
- Builders in Residence: https://www.mozilla.ai/company/builders-in-residence

## Copyright

Portions of content are ©1998-2023 by individual mozilla.org contributors

_Last updated: 2026-02-06_