Your AI gets your company's rules.

The problem appearing everywhere right now

In most companies several people now use AI tools, and each of them explains to their own assistant how the place works. What comes out is as consistent as any one person's memory: one knows the current procedure, another a version from two years ago, a third invented something plausible. The more employees, the wider the drift.

Keeping that aligned by hand does not work in practice. Pasting rules into a chat window creates a copy the moment it is made — and a stale one as soon as the original changes.

One release process, for people and machines

easo did not have to build anything new for this. The ceremony that makes a QMS audit-proof — review, four-eyes release, versioning, signature — is the same one that makes an answer trustworthy. What your team reads in the handbook is what your people's AI is told: the released version, not the draft and not the recollection.

How the access works: through an open standard

easo ships a small server that runs on your machine and talks to the AI tool over MCP — the Model Context Protocol, an open standard for how an assistant asks for data. That sounds like a detail for IT and is in truth a decision about dependence.

A standard rather than one integration per vendor. We did not connect one particular AI tool; we speak the language it speaks — and that others speak too. Change your tool and your QMS stays connected. The same logic as your documents in your own git: we build no door only we can open.

The standard can run locally. MCP has a path where the server and the tool talk on the same machine. That is exactly why our channel needs no network connection. A vendor's own cloud integration could not do this — it would have to send your content through someone else's servers. How MCP works →

What we explicitly do not claim

easo transmits nothing — your AI tool does. The channel reads locally, and that is where our responsibility ends. Where your assistant sends what it read afterwards is decided by that tool and its provider. We tell you what the assistant gets to see, not what it does with it.

The switch governs easo's channel, not file access. Your QMS deliberately consists of open text files in your own git — which is exactly why you can never be locked out. Anyone with access to those files can point any tool at them. What we govern is the official, verified path with provenance attached, and that is a different thing from copy protection.

And we monitor nobody. This is about what the tools are told, never about what employees type.

Why not simply use one of the large platforms

That an AI should answer only from reviewed content is becoming self-evident — there are several providers for it by now, and that is a good thing. The open question is a different one: where does the approved version live, and who can verify it without the vendor?

Platforms that must first collect your company knowledge in order to show it to your AI answer that question with "here, with us". Tools that govern which server an assistant may call do not answer it at all — access control says nothing about whether the content is approved. With easo the source lives in your own git, the release is a cryptographic signature rather than a tick in a workflow, and both can be checked without us. Leaving easo costs you nothing.

Honestly placed

In everyday use the channel reads, and it reads released documents only. Since version 0.10.0 two things have joined it that we promised here once they existed. First, your assistant answers which clauses a text addresses and what is missing for one — it does the reasoning, easo supplies the verified briefing; easo still runs no AI of its own for this. Second, it writes with you, but only inside a session you open yourself and which expires after eight hours.

What holds there, unchanged: it writes only into your own drafts — a colleague's draft is refused, and taking one over or opening a new version stay acts in easo. Every write is one signed commit under your device, attributed to you exactly like your own editing. What the assistant does not know it marks as an open point instead of inventing it. And releasing, signing, withdrawing and everything touching control stay out of its reach.

On a single device, read access is part of the free tier; the company-wide policy with automatic rollout across the team belongs to the Team plan.