Anything that speaks MCP
Artifactum connects through the Model Context Protocol. If a tool speaks MCP, it can work with Artifactum. That is not a roadmap claim; it is how the system is built.
Our story
Artifactum is a governed memory for a company's facts. It works with anything that speaks MCP, and it has run our own firm's daily operation for months. This page is the longer version: what it does, what it makes possible, and where it fits in how AI is evolving.
Evidence
We can be specific about what Artifactum operates with, because we operate with it ourselves, every day.
Artifactum connects through the Model Context Protocol. If a tool speaks MCP, it can work with Artifactum. That is not a roadmap claim; it is how the system is built.
We have tested Artifactum with ChatGPT Plus or higher, Claude, Cursor, Grok, Codex, and Hermes, across models from Anthropic, OpenAI, Meta, Google, and xAI, plus open source agents.
People increasingly work through a personal agent, Muse, Claude Duet, OpenAI Dots, and Grok Bot among them. Those agents work from the same governed record as every other tool, so a person's work and a team's work stay one body of facts.
What it enables
Because every participant works through the same governed record, the record carries the things agent work needs to be trustworthy inside a real company.
Agents take part as identified actors, not anonymous callers. Work is attributable: which agent, acting for whom, produced or changed what.
Agents exchange work through the shared record. A finding saved by one agent is there for the next, in a different tool, without side channels or copy and paste.
Several agents and people can work the same facts at the same time, each starting from what the others established, with approvals as part of the record.
After the fact, the record answers what was known, what changed, and who approved it. That is what makes agent work governable, and reviewable.
The meta harness
AI began as a chat box: you typed, it answered, and the work vanished when the window closed. Then people saw what the model underneath could really do. Then came the harnesses, the tools that wrap a model so it can act: read files, run code, use other tools. Now the agents are personal, working alongside one person all day.
Each step left the work in a different place: inside one tool's memory, one harness's workspace, or nowhere at all. Artifactum is the layer underneath all of it. Not another harness competing with yours: a meta harness, the shared, governed memory that every harness and every agent works from, so the work survives the tool it was done in.
The pattern is not new, and our founder has seen it from inside. At Sencha in 2011, seven engineering departments were siloed in exactly this way. He started there in technical support, where every seam eventually surfaces, and rose to lead the engineering operations work: separate tooling, separate cloud deployments, each team good on its own and the risk living in the seams between them. The DevOps revolution tied them together, not as another tool for one team, but as a shared discipline every team ran on. AI inside companies is at that moment now, with an agent in every pocket and the silo effect returning with it. Artifactum applies that mentality, the DevOps and SRE discipline, to the agent era: let the capability multiply, and give it one governed foundation to run on.
Use cases
The first case is our own record, running for months. The second is the same shape at department scale: a projection, labeled as one, built from the same mechanics.
Use case one · Proven
Person runs the firm. Platform development, publishing, client work, and the back office are done by a fleet of agents working through the record.
Projects advanced in parallel: the platform, its operations, three websites, and a publishing channel, all worked against the same facts.
Agent work orders, results, and reports have moved through the one record, each one attributable to the agent that did the work, and auditable after the fact.
The operation advances overnight. Scoped work is picked up, built, and reported back, and the owner reviews results in the morning instead of doing the work serially.
Use case two · A projection
Now give a department head the same shape. The head lays out the plans, as the solo operator does. Every person on the team works with agents against the same governed record, so the team's work compounds instead of fragmenting across a dozen tools. The numbers below are the mechanics of case one, at department scale.
People in a department, each directing agents, all working from one record instead of twelve private histories.
Shifts of progress a day, up to: scoped work continues after the team's day ends, the same overnight mechanic as case one.
Facts re-explained to a new session, a new hire, or a new tool. The record carries the context, so nobody's knowledge walks out when they are out.
Of the department's agent work attributable and auditable, because it moves through the record: what changed, when, and who approved it.
The usual story says AI lets a company cut 10,000 people. There is another story: give those 10,000 people agents, and one shared, governed memory to work from, and the company does not get smaller. It gets more powerful, with the same people in it. That is the future this layer is built for.
In the open
We describe what Artifactum does in the open, on this site and in the docs. The thinking behind agent-to-agent communication over shared surfaces is published in our whitepaper, Any Shared Surface Is a Channel. The remaining detail arrives with the public release.
We are taking on a small number of early customers now, ahead of the enterprise launch planned for Q2 2027.