Delentia Labs vs Bare LLM APIs
A bare LLM API is a model. Delentia Labs is a constitutional AI operating system. The difference is not performance — it is governance, compliance, and memory.
Bare LLM APIs (OpenAI, Claude, Gemini, etc.)
- • Raw access to one model at a time
- • No persistent memory — context window only
- • Probabilistic output — model may hallucinate
- • No audit trail, no right-to-erasure mechanism
- • Compliance is entirely your job (it is, with Delentia too)
- • No agent-to-agent communication standard
- • Vendor lock-in — switch provider = rewrite code
- • Cost: always paying premium model rates
Best for: rapid prototyping, research, simple tasks
Delentia Labs (Constitutional AI OS)
- • Open-source runtime over a model you choose
- • Persistent memory as a hash-chained event log
- • Deterministic safety (FDIA kill switch: A=0 -> F=0)
- • Signed, hash-chained audit log
- • Erasure of one person's memory text by destroying its key
- • JITNA Protocol for agent-to-agent orchestration
- • Vendor-neutral — swap models without code changes
- • Per-episode cost limit (no saving claimed)
Best for: enterprise production, regulated industries
Feature Comparison Table
| Capability | Bare LLM API | Delentia Labs |
|---|---|---|
| Safety gate before tool calls (in code) | ||
| Signed, hash-chained audit log | ||
| Jury of independent models (optional; scripted tests only) | ||
| Persistent memory across sessions | ||
| Memory erasure for one person that can be verified | ||
| Constitutional kill switch (A=0 -> F=0) | ||
| Thai-language model support (a default table entry, not compared for quality) | ||
| Vendor-neutral (swap models freely) | ||
| Per-episode cost limit | ||
| Hosted, managed service | ||
| Direct API access (no middleware) | ||
| Open source components (Apache 2.0) | ||
| Custom fine-tuning support |
When to Use Which Approach
Enterprise regulated AI (PDPA, finance, healthcare)
Requires audit trail, right to erasure, and Section 33 explainability — all provided by Delentia RCTDB + JITNA.
Quick prototyping / developer experiments
Raw LLM API is simpler for one-off experiments. Delentia OS is designed for production-grade deployments.
Agentic workflows with multiple AI agents
JITNA Protocol provides standardized agent negotiation. LLM APIs have no concept of multi-agent orchestration.
Thai enterprise AI deployment
You can choose a Thai model or one on your own machine, and a data-residency policy is checked before every model call. Delentia does not certify PDPA compliance. It is built in Thailand by one person.
Research / academic use
Both work. Delentia adds reproducibility via signed, verifiable outputs (SignedAI + RCTDB).
How Delentia chooses and switches models
The model is a setting; the safety gate sits outside it