Axiom Works — We build software with AI, not just about AI

Product Portfolio

Four products. Building in the open.

Click any card to see the full picture: capabilities, tech stack, roadmap.

Published

GreyMatter Solo

Your AI forgets everything after every conversation. GreyMatter doesn't. It remembers what you worked on, what you learned, and what matters — and it gets smarter the more you use it. Like having a second brain that actually works.

BSL-1.1 Python Go MCP

Click to expand →

Capabilities

An intelligent harness that wraps your AI coding agent with persistent memory
A second brain that remembers decisions, patterns, and context across every session
Spaced repetition (FSRS-6) that surfaces knowledge before you forget it
Works with Claude, Codex, and Gemini through the MCP protocol
Local LLM inference via Ollama — works fully offline, no cloud required
Security scanning on every interaction via SecureLLM

Tech Stack

Python — core runtime and MCP server
SQLite — local knowledge graph and observation store
MCP (Model Context Protocol) — 17 tools
Ollama — local LLM inference for quality gates
Claude Opus, Sonnet, Haiku — primary frontier LLMs
macOS (launchd) and Linux (systemd) service support

What It Solves

Your AI starts from zero every conversation — Solo makes it remember
Corrections you give today are forgotten tomorrow — Solo makes them stick
Your code, your data, your context never leaves your machine
No internet? No problem. Full functionality offline with local LLMs
One install, one command — works in minutes, not days

Roadmap

Q1 '26

Knowledge graph, MCP tools, Ollama integration, installer

Q2 '26

Go coordinator, session dispatch, PyPI publish

Q3 '26

Voice integration, multi-LLM support

Q4 '26

iOS/tvOS app, next-gen IDE

GA Q4 2026

GreyMatter Teams

Your team's collective intelligence, always available. Every person keeps their own private brain. When they share knowledge with the team, it's encrypted, controlled, and retractable. No one loses ownership of their ideas.

BSL-1.1 Python Go mTLS

Click to expand →

Capabilities

Every team member runs their own GreyMatter Solo — their private second brain
Share knowledge across the team with encryption and access controls
Retractable sharing — revoke access to anything you've shared, at any time
Quality gates that review AI-generated code before it merges
A fleet of specialist AI agents, each focused on a language or domain
Live dashboard with cluster topology and agent status

Tech Stack

Go — coordinator, dispatch engine, API server
Python — MCP server, knowledge graph, spaced repetition
SQLite — local persistence on every node
mTLS — mutual TLS on all inter-node communication
Raft consensus for high-availability coordination
macOS, Linux, and air-gapped deployment support

What It Solves

AI agents on your team can't share what they've learned with each other
Sharing knowledge today means giving up control of it — Teams doesn't
No governance exists for fleets of AI coding agents
Enterprise environments need encrypted, auditable AI operations

Roadmap

Q1 '26

mTLS cluster, OTEL tracing, knowledge sync

Q2 '26

Go coordinator, agent dispatch, quality gates

Q3 '26

Multi-cell federation, enterprise features

Q4 '26

Kubernetes operators, enterprise dashboard

GA Q3 2026

SecureLLM

You're already using AI at work. So is everyone on your team. SecureLLM makes sure nobody accidentally shares a social security number, an API key, or a customer's private data with an AI. It sits between your people and the AI, and it catches what humans miss.

BSL-1.1 Rust Tokio

Click to expand →

Capabilities

Catches PII before it reaches the LLM — SSNs, credit cards, API keys, emails
Blocks prompt injection attempts before they hit the model
Classifies content safety on both requests and responses
Works with any LLM provider — Claude, GPT, Ollama, any OpenAI-compatible API
Deploys as an inline proxy or sidecar — no code changes needed
Full audit trail of every AI interaction

Tech Stack

Rust — 14 crates, 26.8K lines of code
Tokio — async I/O for high-throughput scanning
Zero runtime dependencies for the core engine
Provider abstraction — plug in any LLM backend
BSL-1.1 — source-available, free for individuals

What It Solves

People paste sensitive data into AI tools without thinking — SecureLLM catches it
Prompt injection is a real attack vector — SecureLLM blocks it before the model sees it
Compliance requires an audit trail for AI use — SecureLLM provides one
You shouldn't have to trust employees to never make a mistake with AI
Works silently in the background — no training, no behavior change required

Roadmap

Q1 '26

Core scanner, PII detection, 4 LLM providers

Q2 '26

Rust refactor, open source preparation

Q3 '26

crates.io publish, documentation

Q4 '26

Enterprise dashboard, policy engine

Published

YouTube MCP

68 MCP tools for YouTube and YouTube Music. Search, playlists, comments, analytics, library management. The most complete YouTube integration for Claude Code.

Apache 2.0 Python PyPI

Click to expand →

Capabilities

68 tools covering the full YouTube Data API v3 — not just transcripts
Full YouTube Music library management — playlists, likes, albums, artists
Search, upload, comment, moderate, analyze — everything the API supports
When the API has gaps, we document them and contribute upstream fixes
OAuth 2.0 with TV client flow for YouTube Music authentication
Works with Claude Code, Codex CLI, Gemini CLI, or any MCP client

Tech Stack

Python — published on PyPI
Apache 2.0 — fully open source, use it however you want
Google YouTube Data API v3
ytmusicapi — unofficial YouTube Music API
MCP (Model Context Protocol) — standard tool interface
Active upstream contributions to ytmusicapi

What It Solves

Every other YouTube MCP server just does transcripts — this one does everything
YouTube Music had no MCP integration at all — now it does
Manage your entire YouTube presence from inside your AI coding agent
API gaps are documented, not hidden — and we fix what we can upstream
One install, one config — pip install and you're running

Roadmap

Q1 '26

68 tools, PyPI publish, YouTube Music support

Q2 '26

Upstream contributions, API gap documentation

Local First

Your data stays on your hardware.

Local First is a set of principles where your device is the primary authority for your data — not a cloud server. Coined by Ink & Switch in 2019, it's how software should work: fast, private, and yours.

Fast

Reads and writes happen against local storage. No spinners, no round trips. Sub-millisecond response times because nothing leaves your machine.

Works Offline

Full read and write capability without internet. Not "gracefully degrades" — actually works. Your AI tools function wherever your laptop does.

Private by Architecture

We don't see your data. Not because of a privacy policy — because the architecture makes it impossible. Your code and context never leave your device.

You Own Your Data

If we disappear tomorrow, your tools and data still work. No sunset, no export deadline, no scramble. Local files, open formats, on your hardware.

Compliance by Default

GDPR, HIPAA, data sovereignty — local-first satisfies them structurally. No DPA needed with a cloud vendor if the data never leaves your jurisdiction.

Cost at Scale

Cloud inference charges per token, forever. Local inference costs hardware once and runs as many times as you need. At scale, local wins.

Part of the Local First movement.

Compliance & Security

Built for environments where compliance isn't optional.

Defense, law enforcement, financial services, healthcare. Our products are being built to meet the standards these industries require.

FIPS 140-3

Post-quantum algorithms implemented (ML-KEM, ML-DSA). Targeting FIPS 140-3 cryptographic module validation.

Algorithms Implemented

CJIS

Criminal Justice Information Services security policy compliance for law enforcement data.

In Progress

Air-Gapped Deployment

Designed for full functionality without internet. Local LLM inference via Ollama, on-premises knowledge graph, offline-first architecture.

Designed

Post-Quantum Encryption

ML-KEM and ML-DSA. Quantum-resistant key exchange and signatures. 26K+ LOC, zero dependencies.

Implemented

SOC 2

Security, availability, and confidentiality controls for enterprise SaaS deployments.

Planned

GDPR

Data residency, consent management, right to deletion. Built into GreyMatter Teams from day one.

In Progress

GovRamp / FedRAMP

Federal cloud authorization framework. Required for government cloud deployments.

Planned

mTLS Everywhere

Mutual TLS on all inter-node communication. Certificate pinning. No plaintext in transit.

Implemented

SIGS

Shared Information Governance Standards for cross-agency data sharing in law enforcement.

Planned

Why Axiom Works

Agentic AI software development. No shortcuts.

Axiom Works is a software development company that builds with AI, not just about AI. One developer and a fleet of AI agents build production software through a self-correcting pipeline with auto-generated architectural rules. Every work item is planned, executed, reviewed, and verified before it merges.

The products are local-first, developer-sovereign, and open source at the core. No cloud dependency. No vendor lock-in. No data leaves your network unless you say so.

Local-First

Your data, your hardware, your rules. Every product runs on-premises or air-gapped. Cloud is optional, never required.

Developer-Sovereign

No vendor lock-in. Open standards, open protocols, open source core. You own the stack.

Compliance-Ready Architecture

Post-quantum cryptography (ML-KEM, ML-DSA). Targeting FIPS 140-3 and CJIS. Built for defense, law enforcement, and regulated industries.

Self-Correcting Pipeline

A fleet of AI agents, quality gates, automated review. Work items flow from plan to production without manual intervention.