OBSERVATORY CLOCK
---- -- -- --:--:-- UTC
4 OBSERVATORIES · LOCAL-FIRST
⟢ AN INTELLIGENCE COMPANY

A shower
of minds,
watched from
home.

Persydos AI builds home-centric, multi-agent systems for people who'd rather their intelligence live close. Swarms that think on the kitchen counter, retrieval that never leaks, a bridge to the cloud that carries no keys. Privacy and security first — sovereignty by design.

DESIGN TARGET
109
target document scale for local-first hybrid indexing
DESIGN TARGET
< 5s
target p50 time to consensus across local swarm peers
DESIGN PRINCIPLE
0
private keys transmitted across external boundaries
ARCHITECTURE
4
core observatories in the Persydos stack
DEFAULT POSTURE
100%
default home state on-device; cloud invoked only on request
⟢ MISSION · WHY WE EXIST

Intelligence is better
when it's closer to home.

We're building Persydos AI to put real intelligence — not a thin client, not a remote API — into homes, small teams, and independent operators.

The cloud has done extraordinary work to make AI possible at scale, and it will keep doing so. Alongside it, we think there's room — and real demand — for intelligence that lives closer to the people using it: on the phone in your hand, the laptop on the desk, the small box in the closet.

Democratizing AI, to us, simply means more places where compute, memory, and keys can comfortably live. Persydos AI complements the cloud rather than replacing it, and reaches for it only when you ask.

Yours.

Your data, your devices, your keys. Always.

Open.

Open weights, open protocols, open seats at the table.

Plural.

Many small minds in concert beat one big mind alone.

⟢ PLATFORM · FOUR OBSERVATORIES

Many instruments.
One shared sky.

P-01 MULTI-AGENT PLATFORM

Boterra

Spawn, orchestrate, and observe constellations of agents that live where you live — on the kitchen counter, in the home server, on the family laptop, and on the cloud when the work demands it. Local and remote agents negotiate as one swarm; sensitive context stays home, raw horsepower comes from the cloud.

TARGET SCALE
1 — 10k
TARGET CONSENSUS
< 5 s
EXECUTION
local-first
P-02 SELF-IMPROVING HOME AGENT

Olxion

A long-running agent that lives on the home network and gets smarter the longer it runs. It learns the rhythms of the household through continuous SFT and RL on its own traces — refining its playbooks nightly, paging a cloud swarm only when the work demands it.

DEFAULT STATE
on-device
NIGHTLY
SFT + RL
MEMORY HORIZON
years
P-03 THE BRIDGE BETWEEN WORLDS

Protosend

One secure bridge between your phone, your home agents, your MCP servers, and the cloud. Carry a swarm in your pocket; let it reach back to the kitchen and out to the rack — with end-to-end attestation, no key ever leaves your devices.

ATTESTATION
owner-approved
KEYS IN TRANSIT
zero
ENDPOINTS
every device
P-04 RETRIEVAL & RAG

FilClade

A retrieval layer designed for swarms — many agents asking many questions at once. Hybrid graph-vector index, sub-100ms target recall at billion-document scale, audit-grade lineage.

TARGET INDEX
109 docs
TARGET RECALL
< 100 ms
LINEAGE
cryptographic
⟢ MANIFESTO · ARCHITECTURAL PRINCIPLE

Sovereignty
is not a
feature.

An intelligence that only works by streaming your keys, your memory, and your private context to someone else's cluster is not yours. We build multi-agent systems that execute on your hardware first, verify work across independent peers, and treat every retrieval, message, and tool call as an auditable contract under your control.