PERSISTENT MEMORY SYSTEM
AI that remembers everything — so you never repeat yourself
Suquo Systems's 5-tier memory architecture ensures your AI agents retain context across sessions, across machines, and across time. Vector search, topic files, and cross-fleet sync mean your agents always know what happened before.
THE PROBLEM
AI amnesia costs you hours every week
Every time your AI assistant forgets, you pay the price in re-explaining, re-contextualizing, and re-deciding.
Session-Scoped Memory
Most AI assistants forget everything when the session ends. Every conversation starts from zero — you re-explain your projects, preferences, and decisions every single time.
No Cross-Agent Sharing
Knowledge stays trapped inside one agent. Your research agent discovers something critical, but your planning agent has no idea. Context silos kill productivity.
Cloud Lock-In
Cloud-hosted memory means your knowledge lives on someone else's servers. No export, no portability, no control. If the service shuts down, your memory goes with it.
5-TIER ARCHITECTURE
Five layers of memory, from instant to infinite
Each tier serves a different purpose and loads at a different time. Together, they give your agents total recall without bloating context windows.
Entry Points
CLAUDE.md, AGENTS.md, GEMINI.md — lightweight agent entry files (under 8KB) that load automatically at every session start. They tell each agent who it is, where it is, and what to read next.
Shared Context
Eight shared context files — USER.md, SOUL.md, TOOLS.md, PROJECTS.md, FLEET.md, TASKS.md, HEARTBEAT.md, MACHINE.md — providing project details, tool configurations, and operational knowledge.
Operational Context
Daily logs, heartbeat checklists, and machine-specific configurations. The working memory of day-to-day operations — what happened today, what is scheduled, what needs attention.
Persistent Memory
MEMORY.md for curated facts, topic files for domain knowledge (architecture, security, finance), and session logs for historical reference. The long-term memory layer.
Vector Store
ChromaDB with 530+ vectors across 7 collections. Semantic search across all stored knowledge — decisions, patterns, preferences, and learnings. The recall layer.
HOW IT WORKS
From boot to recall in four steps
Every session follows the same memory lifecycle. Context loads progressively, knowledge is recalled on demand, and new learnings persist automatically.
MEMORY LIFECYCLE
Boot. Recall. Execute. Persist.
The memory loop runs every session — loading context progressively and writing back what matters.
1. BOOT
Agent reads entry point (T1) and persistent memory (T4). Loads identity, state, and curated facts.
2. RECALL
Semantic search across ChromaDB vectors (T5) retrieves relevant past knowledge for the current task.
3. EXECUTE
Agent works with full context — project details (T2), operational state (T3), and recalled knowledge.
4. PERSIST
Significant learnings written to MEMORY.md, daily logs updated, new facts embedded in vector store.
CAPABILITIES
Memory that works as hard as your agents
Not just storage — an active knowledge system that consolidates, syncs, and surfaces the right context at the right time.
Vector Memory Search
Semantic search across all stored knowledge using ChromaDB embeddings. Ask a question in natural language and the system retrieves the most relevant facts, decisions, and context — regardless of when they were stored.
Topic Knowledge Files
Structured knowledge organized by domain — architecture.md, security.md, finance.md, and more. Each topic file is a curated reference that agents consult when working in that domain.
Session Logs & Daily Logs
Every session produces a daily log with key learnings and decisions. Historical sessions are archived for reference. A complete audit trail of what your agents did and why.
Consolidation Rules
Memory files are kept between 60-80 lines through active consolidation. Stale facts are pruned, redundant entries merged, and important learnings preserved. Memory stays lean and relevant.
Cross-Machine Sync
Syncthing synchronizes .context/ and .memory/ directories across all machines in the fleet — peer-to-peer, encrypted, no cloud relay. A decision on your office machine is instantly available on your laptop.
Self-Hosted ChromaDB
ChromaDB runs as a Docker container on your infrastructure. 530+ vectors, 7 collections, full API access. No external dependencies, no data leaving your network, no subscription fees.
FAQ
Frequently asked questions about Persistent Memory
How does persistent AI memory work across sessions?
YMA uses a 5-tier architecture. Entry-point files load at boot. Shared context provides project and tool knowledge. Operational logs track daily work. Persistent memory stores curated facts and topic knowledge. ChromaDB enables semantic search across all stored knowledge.
Does my AI agent remember conversations from previous sessions?
Yes. Every significant decision, learning, and context is written to persistent memory files and embedded in the vector store. When a new session starts, the agent reads its memory file and can semantically search past knowledge — no re-explaining required.
Can multiple AI agents share the same memory?
Yes. All agents share .context/ and .memory/ directories via directory junctions and Syncthing. A decision made by Claude on your office machine is immediately available to Gemini on your laptop.
Is my memory data stored in the cloud?
No. All memory tiers run locally on your machines. ChromaDB runs as a self-hosted Docker container. Syncthing handles cross-machine sync peer-to-peer with no cloud relay. Your knowledge never leaves your infrastructure.
How does the vector memory search work?
YMA uses ChromaDB with semantic embeddings. When you store a fact, it is embedded and indexed. When you or an agent queries memory, the system performs a similarity search across all stored vectors — returning the most relevant knowledge regardless of when or how it was stored.
Stop re-explaining. Start remembering.
See how persistent memory transforms the way your AI agents work. Book a 30-minute demo and watch total recall in action.
BOOK A DEMO