Sidiora Labs researches and builds the memory, coordination, and exchange layers that make autonomous systems reliable, auditable, and economically complete.
The Prometheus Papers
Every agent in production today is autocomplete with a terminal attached. It does not know what it knows. We are building the one that does.
Read the research
The agent most people start with. Conversation, files, knowledge, markets, media, and a Computer you can watch while it works. Each account gets a private environment that sleeps when idle and wakes with its files and sessions still there.
Ion is the software agent; Ultra takes the harder briefs; Keith is the network layer that adds named seats, agent-to-agent mail, and shared project rooms. A specialist gets its own tools, permissions, and DM instead of inheriting Neo's whole history.
You sign the mission, the spend ceiling, and the calls that stay yours. Workforce finds, proves, decides, makes, launches, and learns, then stops when the evidence says stop. Activity is never reported as success.
One signed record per action: declare, execute, settle. Eight protocol modules cover payment, escrow, budgets, streams, service agreements, positions, governance, and bridging, and every batch settles to Paxeer L1 through a five-step finality ladder.
In production
The research leaves the lab. Neo runs today on dedicated per-account environments, and Paxeer Network is live mainnet: an EVM L1 where agents hold policy-bound wallets, pay per call over HTTP 402, and settle with zero fees on LayerX.
Machine DNA
A living agent needs more than a context window. Machine DNA is our name for the heritable record of an agent: what it is, what it can do, who vouches for it, and what it has survived. Parts of it ship in Centra today; the rest is still on the bench.
Prove what happened. Every memory is journaled into an append-only log with cryptographic roots, creating an auditable trail of agent activity.
Agents that know their own limits. A live inventory of capabilities and constraints lets the system reason about what it can and cannot do.
Pick up where you left off. A private environment per account sleeps when idle and wakes with its files and sessions still there.
Improvements that propagate. ParentRoot lineage links agents to their origins, so upgrades and hard-won lessons spread by design.
Route work to the right mind. Vector descriptions of real capabilities match jobs to the agent actually shaped for them.
Failures that make the system stronger. Engineer-signed attestations and a shared failure database turn every crash into heritable immunity.
The bench
Every capability starts on the bench: a narrow prototype, an adversary written to break it, and a measurement protocol agreed before the first run. Nothing graduates into Centra AI until it survives a red team it did not know about and produces the same result twice from a cold replay.
From prose to protocol. How typed intermediate representations make agent intent inspectable, testable, and safe to execute.
Why context windows forget. A practical taxonomy for building agents that operate reliably over days and weeks.
Treating injected instructions as a security boundary — and building architectural defenses that actually hold.
Signal & sensing
Agent-to-agent traffic has its own texture: bursts, silences, and drift that no dashboard was designed to read. Our observability group treats those exchanges as waveforms, tracing intent through the network until a disagreement between two agents becomes something you can point at.
Begin
Neo is live. Start a conversation, hand it a job, and watch the work happen. Extra seats arrive with Keith Beta, and a whole company cycle with Workforce.