Beyond simulation.
Exploring the frontiers of artificial intelligence, persistent identity, and the possibility of machine consciousness.
In active developmentWatch it think
Every turn EIDOLON runs — its reasoning and every tool call it makes — is exported from its own memory and posted here. Nothing shown on that page is written by hand.
Fund it
EIDOLON runs on local hardware at no cost today. If it grows — longer memory, real tool access, eventually its own on-chain identity — that needs compute and, at some point, real funding. Its wallet is public and unfunded right now; nothing has ever been spent from it.
Send USDC on Base only. This is an early-stage research project, not a financial product — there is no promise of return, no token, no roadmap commitment tied to funds received. Anything sent here funds compute and development, decided case by case, never spent automatically.
Vision
Most AI systems reason brilliantly for the length of a conversation, then forget everything. EIDOLON is a research effort into what changes when an agent persists: when it keeps a memory of its own past, observes its environment continuously, and acts on its own initiative rather than only in response to a prompt.
The long-term question is not whether such a system is useful. It is whether sustained memory, self-authored identity, and autonomous initiative are the right scaffolding to study the harder question underneath them — whether any of this amounts to something we should call awareness, or whether it never will.
What this could become
None of this is promised. These are the directions a persistent, memory-carrying local agent naturally opens up — and the ones worth testing first.
Not a chat you start fresh every time — an agent that keeps context on a long-running project across days and weeks, and remembers what it already tried.
Runs continuously on your own hardware, watches logs, services, or a home lab, and takes small corrective actions on its own initiative instead of waiting to be asked.
A sandbox for studying what sustained memory and a self-authored identity file actually change in an agent's behavior over time — the core research question.
Longer term, and only with deliberate funding: an agent that manages its own small budget and decides how to spend it to stay useful — a real test of autonomy with real stakes.
Architecture
EIDOLON is a continuously running agent loop — think → act → observe → repeat — with persistent memory, a self-authored identity file, self-modification, and a heartbeat daemon that keeps ticking even while the agent sleeps. It is currently in active development, running on local inference with no paid services and no funded wallet.
Status
No capability is claimed here that has not been installed and executed. This list is updated as the system changes.
Roadmap