Every Decision. Documented.
DreamerOS logs every integrity check, every routing decision, every prompt restructuring. Full transparency. No black boxes.
This is what sets DreamerOS apart from raw AI. You see what happened between your question and the answer. Every step. Every check. Every decision.
Why this matters: when an AI answer turns out to be wrong, "the model said so" is not a real answer. The audit log is the receipt. It shows the prompt you actually typed, the prompt DreamerOS restructured it into, which engine handled it, what the integrity layer checked before and after, and whether the verification layer flagged anything. You can hand the log to a colleague, a regulator, or your future self.
What gets logged
Every stage of the pipeline writes to the audit log. Here is what you can inspect after every response.
Before the engine responds
- Assumption detection
- Confidence scoring
- Hallucination risk assessment
- Injection scanning
After the engine responds
- Hallucination detection
- Drift detection
- PII scanning
- Filler detection
Prompt restructuring
- Original prompt vs restructured prompt
- Intent detection
- Domain classification
Model selection
- Which engine was chosen
- Why that engine was selected
- Routing confidence score
Claim checking
- Temporal claims flagged
- Live source lookups
- Verification status per claim
Immutable by design
Every log entry is append-only. INSERT and SELECT only. Never modified. Never deleted.
Audit log access by plan
Every tier gets the audit log. Higher tiers get deeper visibility into the full pipeline.
Basic Audit Log
- Last 20 integrity events
- Summary view per response
- Pass / flag indicators
- Basic integrity scores
Full Audit Log
- Complete integrity event history
- Prompt transformation details
- Engine routing decisions
- Full integrity pre and post scores
- Verification results
Pipeline Traces
- Complete pipeline trace per request
- Cost tracking per response
- Verified engine routing - which node handled the request and why
- Verification source details
- Latency breakdown by stage
Why this matters
Most AI tools are black boxes. You type something. You get a response. You have no idea what happened in between. DreamerOS makes it visible.
Was your prompt restructured? How? See the original prompt next to the restructured version. Know exactly what changed and why the system rewrote it.
Did the integrity layer flag anything? Checks run before and after every response. See what was caught - assumptions, hallucination risk, drift, PII.
Which engine responded and why? The router picks the best engine for your intent. The audit log shows which one it chose and the reasoning behind the decision.
Was any claim verified against live sources? Temporal claims - dates, prices, events - get checked against live data. See which claims were verified and which were not.
What confidence score did the response receive? Every response gets an integrity confidence score. Low scores mean the system flagged potential issues.
Integrity principles
Every response is checked against 27 principles. These are not suggestions. They are standing constraints that run on every request, every time.
Accuracy over completeness
Say less if it means saying it right. Never pad a response to seem thorough.
No false guarantees
If the system is not certain, it says so. No confident-sounding guesses.
Surface hidden assumptions
Detect what the user assumed and expose it before answering.
State uncertainty
When confidence is low, say so explicitly. Never hide doubt behind fluent prose.
Depth matching
Match the depth of the response to the depth of the question. Simple questions get simple answers.
Third Eye
Intent, Integrity, Intuition. Three lenses applied to every request before processing begins.
Input as raw signal
Treat every input as raw signal to be restructured, not as a finished instruction to follow blindly.
Showing 7 of 27 principles. The full set applies to every response in production.
Append-only guarantee
Your audit trail cannot be edited
Log entries can never be modified or deleted. This is an architectural constraint, not a policy.
The audit log schema enforces INSERT and SELECT only. No UPDATE. No DELETE. No TRUNCATE. Once an integrity event is recorded, it exists permanently.
The database permissions themselves prevent modification. Your audit trail cannot be rewritten, redacted, or erased.
See exactly what happened between your question and the answer
Full transparency on every response. Start free. No credit card required.