Truthound AI¶
Truthound AI is an additive review-layer namespace on top of the Truthound
validation kernel. It does not replace th.check(), ValidationRunResult, or
the deterministic suite/runtime model. Instead, it turns prompts and run
evidence into reviewable artifacts that a human can inspect, approve, reject,
or apply.
Design Intent¶
The AI surface is deliberately narrow:
- proposal compilation: prompt to persisted suite proposal artifact
- run analysis: validation run to persisted operational analysis artifact
- approval and apply: explicit human-reviewed state transitions
- provider and privacy boundaries: summary-only, reviewable, and testable
The goal is not autonomous mutation. The goal is controlled operational assistance that stays subordinate to the core validation contract.
Truthound 3.1.3 also keeps the hardened prompt path for Korean, English, and mixed requests. Natural-language prompts are normalized into canonical validation intent candidates, provider output is checked through structured contracts, and unsupported or ambiguous prompts become reviewable rejected items instead of route failures.
Public Entry Points¶
Install the additive namespace with:
Then use the root probe and AI namespace:
import truthound as th
import truthound.ai as thai
status = th.get_ai_support_status()
if status.ready:
proposal = thai.suggest_suite(prompt="Require customer_id to be unique")
Key public entry points:
truthound.has_ai_support()truthound.get_ai_support_status()truthound.ai.suggest_suite(...)truthound.ai.explain_run(...)truthound.ai.approve_proposal(...)truthound.ai.reject_proposal(...)truthound.ai.apply_proposal(...)
Documentation Map¶
- System Boundary
- Proposal Compiler
- Run Analysis Evidence Model
- Artifact Schema
- Approval and Apply Semantics
- Privacy and Redaction
- Provider Contract
- Prompt Hardening
- Smoke and Release Gates