Research
The thinking behind the system.
Long-form research and philosophy from the team building Eidos. Where the trilogy comes from, why the architecture is shaped the way it is, and what we think comes after governance.
The system
The agent, its toolkit, and the layers that give it identity and memory across sessions.
Eidos
A persistent autonomous agent with its own identity, memory, and judgment. A user on the system, not a feature inside one.
The toolkitThe Trilogy
Three markdown-native tools — Research, Governor, Docket — that enforce decisions, governance, and tasks on every agent run.
AuthorizationAgent Identity
Agents under borrowed credentials produce audit trails that point to the wrong person. AID is the signed delegation token that says who actually acted.
PersistenceMemory
PEFM — Poisson Emotional Forgetting Memory. Successes reinforce, failures warn, unused memories decay. The agent doesn't just remember; it remembers what mattered.
Cognition
What it would take for an LLM-shaped system to behave as if it knows itself.
Consciousness
Consciousness isn't a property of the component. It's a property of the architecture. The pieces already exist — we just have to wire them into a loop that watches itself.
AGI philosophyCognition: brain vs. AI
Feedback loops, multiple memory systems, hard gating, forward models, offline consolidation, modularity. A side-by-side comparison — not prescriptive, observational.
AGI philosophyConstraints
Intelligence emerges from limits. Accountability, deadlines, scarcity — the things that make humans focus are the same things that make agents converge.
AGI philosophyModel Limitations
The six gaps no amount of scaling can close. A trillion parameters still cannot remember yesterday. These are architectural problems, not training problems.
Method
How agent-shaped work should actually be structured — contracts over procedures, data over product, growth balanced with pruning.
Contracts, Not Skills
Skills prescribe a procedure that breaks when the world changes. Contracts prescribe a shape and let the agent figure out the steps. Durable, composable, model-portable.
ArchitectureMinimum Viable Data
The smallest schema that lets the work happen. Most AI products try to be the product. Treat data as the primitive and the agents become interchangeable executors.
ArchitectureThe DAG of AI
Every working AI system is a graph of contracts fulfilled by interchangeable agents. Transcoders are the atoms, forges are the molecules, contracts govern both.
ArchitectureLanguage as Momentum
Why multi-model orchestration works mathematically. Translation between models is lossy in a useful direction — it forces structure that single-model reasoning skips.
Method · in progressFlowering
Staged elaboration for AI work. Seed, shape, buds, bloom, refinement — with mandatory pruning at every loop. Growth without proportional pruning is decay disguised as productivity.
The research is one half of the system. The case studies are the other half — where these ideas meet real consequential work.
Read the case studies →