// memory active

AtomCorp

Reusable Runtime Infrastructure

Your AI should remember you. Z1 does.

⚙ Reserve Your Silo
$20/silo/month · Free tier live now · You keep your silos if you cancel.
last Tuesday you mentioned the project was close you prefer directness you've asked about this before context restored from 14 days ago you said: don't let me forget this recalling a conversation from three sessions ago this pattern again — same as last month silo RELATIONAL: warm start last Tuesday you mentioned the project was close you prefer directness you've asked about this before context restored from 14 days ago you said: don't let me forget this recalling a conversation from three sessions ago this pattern again — same as last month silo RELATIONAL: warm start
The Model Proposes. Python Disposes.
// Memory, reflection, and compression run in deterministic code — so the model just reasons, and never has to remember to remember.
The Memory Stack
🔒
// rmpl_core.py
Runtime Memory Persistence Ledger
The spine. Every memory write is tracked, versioned, and provenance-logged before it touches storage — so nothing gets lost, and nothing gets silently overwritten.
100% context portability — 10/10 critical facts preserved across sessions
🌊
// z1_dam.py
Dam Layer
First gate. Stale, conflicting, or low-value context gets filtered before it's ever saved — so your silos stay lean and recall stays fast.
100% catch rate — 28/28 cases
🏺
// z1_reservoir_gate.py
Reservoir Gate
Where memory settles. Reflected, compressed, and staged before it's written long-term — so history keeps growing without recall slowing down.
Runs automatically after every session — no manual triage.
🌉
// z1_bridge.py
The Bridge
The runtime interface your agent talks to. One connection point for read/write memory — plug in MCP, a custom tool, or your own agent loop.
Framework-agnostic — build whatever you want on top.
🗂️
// z1_silo_router.py
Silo Router
Deterministic keyword matching routes context to the correct memory silo — operational, relational, domain, situational. No model inference for routing, no wasted tokens.
100% routing accuracy — 45/45 cases
🛡️
// boundary safety
Boundary Safety
Runs quietly in the background. Deterministic rules catch destructive or looping actions before they execute; a small local model (llama3.2:3b) handles the fuzzy edge cases. It's there so you don't have to think about it.
Present. Never the headline.
Inside the Silos
Operational
click to open
Day-to-day working state — current tasks, recent actions, what's in motion right now. Short memory, high turnover, always current.
Relational
click to open
Who you are to it. Preferences, tone, the shape of how you like to work. The part that makes it feel like it knows you.
Domain
click to open
Durable facts about your world — projects, terminology, structure. Slow to change. Fast to recall.
Situational
click to open
The moment. What just happened, what's happening now, what's likely next. Short half-life, high relevance while it lasts.
It Remembers
⚙ Speak, and it Will Keep
Operational
Relational
Domain
Situational
vs. RAG & MCP
CapabilityRAGMCPZ1 / AtomCorp
Cross-session memory Persistent across sessions
Categorized context silos Not one giant blob
Provenance tracking Every write logged
Deterministic compression & reflection Runs off the model
Token efficiency at scaleDegradesVery LowImproves
Local / offline operation Zero cloud dependency
Plug into MCP / your own toolsPartial Framework-agnostic runtime
Built-in safety net Present, quietly
"They're trying to pour the ocean through a cup. I built the reservoir."
— Adam Dolin, Founder · NSF REACH Scholar · adam@atomcorp.ai · 623-217-5306
The Factory Is Running.
Proof of concept validated. Loop closed. Phase 2 in progress.
If you're building agents, tools, or MCP servers and you're tired of rebuilding context every session — talk to us.
⚙ See Products Request Access
// or drop your email — we'll reach out
0 context turns persisted · 4 silos active · 0 memories compressed