The runtime that makes AI remember, and stay in bounds
Yohanun sits between your application and any LLM: OpenAI, Anthropic, Google, DeepSeek, Mistral, or a model you run yourself. It's what lets you hand AI real responsibility (client files, personal history, actions with consequences) because nothing gets recalled, revealed, or done that the platform hasn't already cleared.
Three components. One semantic runtime.
Yohanun provides the essential infrastructure that every intelligent system needs: memory that persists, rules that explain, and context that evolves.
Memory Layer
Hybrid vector + graph storage that survives restarts, learns from patterns, and builds semantic understanding over time.
Rules Engine
Declarative business logic that creates explainable, auditable decisions. Define rules once, trust them everywhere.
Context Manager
Intelligent context weaving that understands what matters when. Right information, right time, every time.
Works with any LLM. No vendor lock-in.
Yohanun doesn't replace your LLM; it makes it smarter. Whether you're using GPT-5, Claude, Gemini, DeepSeek, or running local LLMs, Yohanun provides the same semantic infrastructure.
Switch models without rebuilding your memory. Change providers without losing context. Build once, run anywhere.
Supported Models
Governed by design. The model is never the gatekeeper.
The result: leaks that can't happen, agents that can't overreach, and an audit trail that answers your compliance team's questions before they're asked. All enforced by the platform, the model is never the gatekeeper.
Clearances
Platform-owned clearance ledger controls who can see what. Labeled memory, conflict-of-interest checks, and instant revocation, enforced on every read.
Mandates
Agents act up to an explicit limit and no further. Anything over the line escalates to a human decision queue. Commits fail closed; every action is audited.
Learning Loop
Report outcomes (success, failure, or correction) and they become retrievable lessons. Memory that helped ranks higher; memory that hurt ranks lower. Routines that keep working are proposed as procedures; a person ratifies them; agents follow them.
Facts, Not Just Text
Conversations and documents yield one-claim facts with names spelled out and dates attached, each anchored to the document, page and section it came from. Search by period, and ask what was true between two dates.
In plain English: a paragraph is a haystack; a fact is a needle with a label on it. We keep both.
A Record Never Rewritten
Memory is closed or superseded, never edited; every change is a ledger row with lineage both ways. A judge proposes duplicates, refinements and contradictions; a person ratifies, rejects, or keeps both. Ask what the system believed on any date.
In plain English: track changes that cannot be switched off, with a human deciding what counts as the truth.
Governed Erasure
Nothing is destroyed by accident. When erasure is required, a purge resolves the target to a plan with no content in it, a mandate or a human approves exactly that plan, every store is scrubbed, and each erased memory leaves a tombstone.
In plain English: destroy it, and prove you did. Both halves, with a receipt.
Simple integration. Powerful results.
Add semantic intelligence to your existing AI workflows in minutes, not months.
# One REST call. Memory, governance, and rules applied by the platform.
curl https://app.yohanun.com/api/ai/chat \
-H "X-API-Key: $YOHANUN_KEY" \
-H "Content-Type: application/json" \
-d '{
"message": "What did we agree with the supplier last month?",
"user_id": "user_123",
"thread_id": "thread_42"
}'
# The response includes the answer, the memory sources it drew on,
# and temporal context. Every source one your principal was cleared to see.
# Streaming (SSE) and WebSocket endpoints take the identical request.Ready to build intelligent systems?
Start on an instance we run for you, in the region you choose. Self-hosting and your own model are available as a premium arrangement.