The operational layer that works with your entire AI stack
Empowering your team to scale AI. Faster, simpler and safer.
(sento) Foundation
The multiplayer context layer shared and shaped by your entire company.
Foundation is a multiplayer context layer: one place for the knowledge, terms, metrics, instructions and skills your company runs on, each with a named owner. It fits the AI stack you already have.
Every serious AI effort hits the same wall. People build skills, workflows and agents, but the knowledge those agents need stays scattered across files, tools and individual heads.
Foundation holds that knowledge once. People and agents work from the same living version of the company.
Write something once and every agent can read it, in whatever tool it runs. The same answer, every time.
Foundation has a fixed structure with strict rules. Every entry is typed, tagged and owned by a person, and it is flagged when it expires.
Each entry carries its own read and write rules for people, agents and scheduled routines. An agent reads what it is allowed to read, a routine writes only where it is allowed to write, and nothing gets in by accident.
Together the entries form a semantic layer. Every term is defined once, and every metric travels with its definition, its source and what counts as good or bad.
It is also where everything built on it is run from: one place to deploy, monitor, debug and improve every skill, routine and agent.
Set up and manage all of it from the AI tool you already use, Claude Code or Codex.
Your AI stack today
Stuck in silos, hard to distribute, impossible to keep in check.
We have supported companies of all sizes in setting up stacks like this, and the same problem showed up every time.
The instructions end up in markdown files. Setting them up is easy. Keeping them right is not: they drift apart, break quietly, and get debugged one tool at a time. And they live wherever one person put them, so they are never reachable from every tool or open to everyone who needs them.
So we built a base the whole organization can work from, not just the one or two people who set things up.
Enabling
A joint surface, under your control.
Multiplayer
You stop rebuilding context
Every new skill or agent starts from the same definitions, metrics and instructions. The work of explaining the company to each new agent goes away.
Semantics
Your terms mean one thing
Every term the company uses is defined once, with what it includes and what it does not. People and agents read the same definition, so a word means the same in every answer.
Ownership
Domain experts own the content
The people who know the work write it. Every entry has a named owner and its own rules, so when something is wrong there is one person to ask, and one person who can fix it.
Access
Reads and writes are limited by design
Each entry can be restricted to specific people and agents, for reading, for writing, or both. Everyone else is kept out.
Independence
Model agnostic by design
You own the context layer, so no model owns you. Claude, ChatGPT, Codex and custom agents read the same version and quote the same numbers, and you can switch providers at any time without losing a thing.
Metrics
Numbers travel with their meaning
A metric carries its definition, its source and what counts as good or bad. An agent quoting it can say what it means and whether it is a problem, not just the figure.
Routines
Routines keep it current
Scheduled routines pull from your other systems and write the result the same way every time. Metrics and lists update without anyone retyping them.
Feedback loop
Missing knowledge becomes visible
When an agent asks for something that is not captured yet, the MCP records the request instead of guessing. Define it, and the layer grows from what people actually need.
Distribution
One place to run it all from
Deploy, monitor, debug and improve every skill, routine and agent from the layer itself, instead of chasing each one where it happens to live.
Logging
Every read and write is logged
See what was read, updated or appended, when, and by whom. When an agent gives an odd answer, you can trace exactly what it was working from.
Versioning
Every update is versioned
Each change creates a new version on its own. Nothing is overwritten, and you can always see what an entry said before.
Hosting
You keep the keys
Run it hosted or on your own servers. The layer is yours and the models stay interchangeable, so switching from one provider to another changes nothing about what your agents know.
Deployment
Hosted by us, or on your own servers. Start alone, then bring the team.
Start on your own
Connect Foundation to Claude Code and try it on your own work before anyone else is involved.
Adding a person is the whole onboarding
They get the instructions and an MCP link. No new portal, nothing to install.
One connection, every client
Claude Code, Claude, ChatGPT and any other MCP client read the same layer.
Hosted by us
A workspace on our infrastructure, ready the same day.
Or run it yourself
Take the code and host it wherever you want. You get two connections: one to your own workspace, and one to Sento's Developer MCP, which delivers updates, guidelines and patches while you adapt the rest in Claude Code.
Unlocked
What others have built around their context layers
A bot in the team's Slack watches for decisions, definitions and numbers as they are agreed in conversation, and files each one as a proposed entry with the person who said it as owner. Nothing lands until that person confirms. The layer grows from where the talking happens, without anyone opening a form.
A set of skills for people starting with AI at the company: what the tools are, how to ask, and what the company already knows. Every exercise reads the same context layer, so the practice uses real terms and real numbers. Progress is tracked per person, so the rollout team can see who is fluent and who is stuck.
One skill reads the changelog, the tone of voice and the customer list from the layer, drafts the release email, and logs what was sent to whom. The changelog, the writing rules and the recipients live in one place, so the whole process runs from there and every send is traceable.
A scheduled agent reads customers, support tickets and usage every morning and writes the accounts that look shaky to the renewal risk list, with the signal behind each one. Before a renewal call, the account owner asks for a brief and gets the same list the agent wrote, not a fresh guess.
Buyers' security reviews are answered from the questionnaire entry: every answer already given once, served the same way every time. A question with no answer yet is recorded as a gap for the security owner to define, so each questionnaire is faster than the last.
A routine pulls MRR, churn, active customers and the quarter's OKRs from the layer, reads what shipped from production pushes, and drafts the update in the company's own voice. The numbers are the ones every other tool serves, and the draft is versioned, so the board reads the same figures the team does.
Journal
Notes from the road as we build Sento.
9 Sept 2026
Making AI multiplayer
3 Sept 2026
We let agents write to our company brain, and we sleep like babies.
28 Aug 2026
Best Company Brain Tools in 2026
27 Aug 2026
Company Memory Should Not Live in Chat
Enable the people who actually know the work. Give every tool and agent the same instructions, definitions, metrics and skills, from one place. We build the layer that holds them, and help companies put it in place. Same context. Every tool.