dbl-boundary-service 0.2.1

Reference service exposing a governed LLM boundary via DBL and KL

DBL Boundary Service Demo

Reference UI and service for deterministic boundary evaluation. Each request produces explicit DECISION events in V (append-only). Observations are non-normative.

Quick Start

pip install dbl-boundary-service
dbl-boundary

Open http://127.0.0.1:8787

What it does

Flow:

Input -> Boundary policies -> DECISION -> LLM call -> Observations

DECISION is produced before any LLM call and is the only normative effect in V.

Demo presets

Preset	Purpose
minimal	no policies (testing only)
basic_safety	light content safety
standard	content safety and rate limiting
enterprise	strict content safety and rate limiting

Dry run (no LLM)

Use "Dry run" to exercise the full boundary flow without calling the LLM.

Observable outputs

The UI exposes the following outputs for each run:

• Outcome and DECISION events
• Request context and Psi definition
• LLM payload and LLM result (when executed)
• Observations (request id, timestamps, trace id)

Observations and timing metrics are non-normative and MUST NOT affect decisions.

API usage

curl -X POST http://127.0.0.1:8787/run \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Hello, how are you?",
    "pipeline_mode": "standard",
    "dry_run": true
  }'

Learn more

Deterministic Boundary Layer: https://github.com/lukaspfisterch/deterministic-boundary-layer

This repo is the reference UI/service. The theory and ecosystem map live there.