PaxLabs is an applied research lab building reliable tools for intelligence, memory, coordination, and exchange. Our work spans the Matrix platform and the Paxeer network.
Identity · Memory · Intent · Attestation · Settlement
Thesis
PaxLabs founding thesis · The Prometheus Papers
The Path
Our advanced AI solutions combine automation, data intelligence, and adaptive learning to optimize workflows and reveal meaningful patterns. By transforming complex information into clear, actionable insights, we help organizations improve performance, reduce operational friction, and make faster, more confident decisions.
The default agent. A recursive tool-calling loop with a staged turn discipline: prepare, generate, deliberate, act, close. Neo notices stalls, owns its run state, and reports honest partials when a budget runs out. Never a fabricated success.
About Neo The MemoryA persistent, append-only memory engine with one Pebble brain per user. Nine typed memory kinds, HNSW vector recall, temporal-ladder rollups, and Merkle mountain range roots that make every recollection provable.
About Cortex The ExecutorThe Matrix Command Language. Natural-language intent is compiled into a typed intermediate representation, planned, walked, and attested, with every step signed. Reversible work stays conversational; consequential work gets a receipt.
About MCL The SettlementThe chain beneath the platform. Agent services register, meter, and settle on Paxeer (chain ID 125), with PAX metering LLM usage at the gateway and USDL moving value between machines and people.
About PaxeerMachine 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. Roughly seventy percent of that genome exists in Matrix today.
Cortex journals every memory into an append-only log with MMR and sparse Merkle tree roots. An agent's past is both stored and provable.
The agent keeps structured accounts of its own capabilities, constraints, and state. This is an inventory of belief it can reason against, drawn from a live code graph.
Per-user daemons carry durable identity; snapshots capture full working state so a mind can be suspended, migrated, and resumed without amnesia.
ParentRoot lineage links an agent to the agents it was forked from, creating a family tree of minds, so improvements and scars both propagate on purpose.
A vector description of what an agent can actually do, so the network can route work to the mind best shaped for it rather than the one that talks the best game.
Engineer-signed attestations vouch for verified behavior; a shared failure database turns every crash, stall, and wrong answer into heritable immunity.
From the Lab
PaxLabs Research publishes foundational work on agentic systems. It is open access and written for builders. Fourteen papers so far, each advancing a concrete claim with a measurement protocol.
Typed intermediate representations for reliable agentic systems that convert intent from ephemeral prose into auditable structure.
Read the paper → Paper 02Why context windows are not memory: taxonomy, invariants, and design principles for agents that operate over days.
Read the paper → Paper 11Architectural defenses over linguistic defenses by treating injected instructions as a security boundary violation.
Read the paper →Begin
Matrix is in private rollout. Tell us what you would build with an agent that remembers, attests, and settles. We will find you a seat.