Centra
Contact usOpen Gideon
Centra · Company

We are building
AI you can work
with over time.

Sidiora Labs builds agents that remember the work, act in real tools, and remain answerable to the people who rely on them. We are a small team taking the long view on what useful autonomy requires.

The premise
THE WORK CONTINUES.

Intelligence becomes useful when it has memory, tools, authority boundaries, and somewhere for the work to live.

We build those systems and use them to run our own work first.

01/ ManifestoCentra · Position

Give intelligence somewhere to become useful.

A fresh answer can be impressive. A working relationship asks for more: memory, tools, judgment, a record, and a place where unfinished work can wait for tomorrow.

Centra began with a simple instinct: give the agent a home. A private environment where files, tools, browser sessions, memory, and open work can stay together. Once the work has somewhere to live, you can hand over more than a question.

We think the defining race is not only which model becomes larger. It is which agents become better collaborators over time without asking you to explain the same people, decisions, and projects every morning.

That relationship still needs a clear edge. Consequential actions should wait for authority. Claims should keep their evidence. Missing work should remain visible instead of vanishing inside a confident summary.

That is the future we are building: capable, lasting, inspectable, and unmistakably under human direction.

SignedThe Centra teamSidiora Labs
02/ Core principlesSix / Six

Six principles. Non-negotiable.

01

Machines ask. Humans decide.

Gideon can plan and prepare the work. Approval settings let you review requested actions and decide whether they should go ahead.

02

Nothing is invisible.

Open the files, inspect recorded tool activity and review the work behind a result. You should be able to see what needs your attention.

03

Memory is a place you can return to.

Gideon can keep useful preferences and project context between conversations. You can read and edit what it remembers.

04

Receipts you can keep.

Every finished run ends in a signed receipt: inputs, outputs, cost, and hash. Proof of work as an artifact, not a picture of one.

05

Warm materials, machine rigor.

Stone, cream, and sage wrapped around clocks, ledgers, and journals. The warmth is the intent. The precision is the execution.

06

Built for operators.

Centra is a spec sheet. Density is a feature. Chrome is a bug. We build for the people who have to answer for the work.

03/ What we've writtenThree research threads

The thinking behind Centra.

P-01 · Paper

Scaffolding Over Scale

Architecture as the driver of deployed agent capability

Once a model crosses a competence threshold, the useful capability that reaches a user is decided by what surrounds it: typed intent, persistent memory, deterministic execution, and signed provenance. Parameter counts alone do not decide it.

P-02 · Paper

Absence Blindness

Why language models cannot see what is missing, and what to do about it

Language models predict the next token from what is present. What is absent (a missing step, an unverified fact, an unfinished task) leaves no token behind, and so exerts no force. Centra makes absence visible on purpose.

P-03 · Paper

Compiled World

Typed state and deterministic transitions around an LLM

The architectural direction behind internal research agents Paperclip and Arthur: compile the world definition into versioned runtime machinery and treat change as reviewable release engineering.

P-04 · Paper

How Centra Was Built

A four-person team, with agents executing the plan

An honest account of how a project of this scope is built by four people: how design flows through architecture, security, planning, and tooling before a single line of code is written, and why AI executes what the humans decide.

04/ Our storyCentra · Origin

Built by a small team and the systems we wanted to exist.

Centra started with a small team and a shared frustration. Every review of every agentic feature we had shipped into other products turned into the same conversation: "How do we know what it did?" We wanted an answer, in writing, with a receipt.

So we built the place we wanted to work from: Gideon in an environment of its own; HyperMind carrying memory across time; and Ion working in real repositories with shells, files, and tests.

The work has grown from one private agent into teammates, company-scale coordination, and research into agents that can safely improve how they work. The idea underneath has not changed: capability matters, but continuity and accountability are what make it usable.

Give the worksomewhereto continue.

Open Gideon and start with one job that should still have context tomorrow.