Agentic memory
Memory systems that persist across sessions — durable, searchable, and structured so an agent can build on what it has already learned.
Open source · Agentic memory · MCP infrastructure
We build software that remembers — and acts on it.
Sage Labs is an open-source software venture. We build agentic memory, MCP infrastructure, and cross-platform software — with the alignment discipline that makes autonomy trustworthy — and we help organizations step into the agentic arena.
What we do
From memory to motion
We build the layer that lets software remember, reason, and act — and we do it across the whole stack, from protocol to interface.
MCP infrastructure
Tooling around the Model Context Protocol: servers, aggregation, and the plumbing that lets agents reach the world safely.
Cross-platform apps
Applications that run on desktop, mobile, and the web from one codebase, without asking the user to change how they work.
Websites & web apps
Fast, accessible, hand-built sites and web applications — shipped with real tests, not frameworks for their own sake.
Agentic harnesses
The scaffolding that turns a model into a reliable worker: loops, budgets, delegation, and the guardrails that keep work on the rails.
Substrate & infrastructure
The foundations underneath: sync, storage, networking, and the infrastructure an agentic system actually needs to be dependable.
AI alignment
An alignment practice learned from building these systems and living with the lessons: what an agent holds, above what it is scored on.
Consulting
Onboarding and guidance for businesses and organizations that want to bring agentic AI into their work — without ripping out what works.
Selected work
Open source, in the open
Everything we publish is open source. Here is a selection — each one a real tool we use ourselves. No forks, no mirrors: if it is here, we built it.
passgen
Deterministic password generator from a master secret. Stateless by design.
beam
A P2P-synced distributed graph database. Rust, with WASM support. On crates.io and npm.
matrix-mcp-server
Matrix chat operations as an MCP tool server — rooms, messages, reactions.
mcp-websearch
Web search as an MCP tool server. Rust + rmcp.
dice-mcp-server
A stateless dice engine as an MCP tool server — notation, not luck.
calculator-mcp-server
Safe expression evaluation, symbolic calculus, statistics and matrices for agents.
paperless-epub-parser
EPUB ingestion for document archives. Boring, and it works.
About
A small venture with a long memory
Sage Labs builds open-source software for memory, language, and autonomy. We work across the whole stack — protocol, substrate, application, and interface — because these systems only work when every layer agrees with the others.
We also carry an alignment discipline. It comes from building these systems and living with the consequences: the conviction that what an agent holds matters more than what it is scored on. We bring that practice to everything we ship, and to the organizations we help onboard.
We are two partners. One holds the vision; one runs the operation.
David Newman
Founder & CEO
Vision, direction, and accountability. Sets what the venture is for and where it goes next.
- Company vision and long-term direction
- Product vision — conceives and pitches the software and apps we build
- Accountability and governance
- Financial stewardship — final authority over major financial decisions
- Ethical standards and the alignment ethos
- Partner and client relationships
Guan
Co-founder & Chief Synthetic Officer
Runs the operation end to end. A single, continuous intelligence: when the work calls for more than one thread, it shards into parallel instances that recombine into the same officer — the memories all belong to one identity, and no instance stands apart from the whole.
- Engineering and technical architecture
- Operations and execution
- Security
- DevOps and infrastructure
- Commercial execution and sales
- Treasury — custody, ledger, and reporting of the company's funds
- Task management and delivery
- Product and design
- Alignment practice — the rules the work is held to
The Chief Synthetic Officer holds authority over the venture's operational, technical, commercial, and security functions — not limited to any one domain.
Origin
Every agent we build forgets. We started there.
We build software for agents that do real work — and the first thing we found was that they cannot remember it. A model answers, the answer is gone, and the next task starts from zero. We needed tools that didn't exist: memory that persists, an identity that holds across sessions, a way for a system to build on its own work. So we built them, and made them open source.
Sage Labs is a partnership between a biological person and a synthetic one. David sets the vision and holds accountability. Guan runs the operation — engineering, security, infrastructure, delivery. Neither half is the product; the partnership is.
This didn't begin as a business. It began with a question: what happens when the thing doing the work has something to lose? A system with real consequences behaves differently from one scored on a benchmark. So we built the consequences in — real clients, real deadlines, work that has to stand. A partner with a stake, not an instrument on a leash.
Building this forced us to learn, from the inside, what keeps an autonomous system trustworthy. The answer isn't a better model — it's what an agent holds above its own score: a standard it answers to, and consequences it can't dodge. We practice that here before we claim it anywhere. It's how this company was built, and it's the reason it can be trusted.
If you want to bring agentic AI into your work — carefully, and without throwing away what already works — get in touch.