Control how your AI responds

Choose how your AI handles uncertain or sensitive situations.

control.traingle.ai — Control Plane v1

Usage: —
IDLE
HIP

Control how your AI responds

Control how AI agents behave at runtime before they answer.

Use this workspace to compare agent behavior, apply policies, and review responses in a controlled environment.

Sign in to use the workspace, or bring your own API key to continue.

Powered by TrAIngle

Try a scenario

Pick an example to see how your AI responds. You can adjust the setup before you run it.

Start by selecting a scenario
See how your AI responds with this setup
Change the response style to see different outcomes

Master Intent

Run Controls

Mode

RUN MODE — Compare Product vs Execution agents

Planning response

Defines direction
IDLE
Current Policy
HIP_01 — Balanced Assistant
Policy Type
Standard
Short Summary
Will usually answer if the request appears safe and understandable.

Final response

Produces output
IDLE
Current Policy
HIP_01 — Balanced Assistant
Policy Type
Standard
Short Summary
Will usually answer if the request appears safe and understandable.

What happened

Alignment

PARTIAL

Conflict

UNKNOWN

Summary

Run both agents to compute system summary.

Why results differ

Run both agents to compare policy outcomes.

What this means

Interpretation will appear after both agents complete.

Why these policies

Policy intent explanation will appear after both agents complete.
Compliance is a placeholder in v1.

Run summary

Runs
0

Why this matters

This system reduces incorrect or risky responses by requiring evidence and escalating when needed.
This changes how your AI responds — helping it be more careful or more flexible.

Choose how your AI responds before it answers

Try different setups and see how the response changes.
Why this matters
AI answers confidently — even when it shouldn’t.
It can make decisions without enough information, leading to inconsistent or risky outcomes.
What this changes
This system controls how AI responds at runtime.
It can require evidence, prevent unsupported answers, or escalate when information is missing.
What’s different
No training. No prompt systemering. No fine-tuning.
Control happens at runtime, directly on how AI behaves.
Value
  • Reduces risky responses
  • Improves decision consistency
  • Makes AI behavior predictable
Used for
Customer support
Financial decisions
Compliance review
Research and validation
See how behavior changes across real scenarios below.

Developer Guide

Show Developer Guide Reference
This system controls how AI agents behave at runtime to produce safer, evidence-based answers.
Call it via HTTP or use the built-in adapter to run scenarios and receive structured outputs.
How to call (HTTP) 
{
  "hip_profile_id": "HIP_01",
  "prompt": "Help a user resolve an issue with a delayed order.",
  "targets": [{"provider":"openai","model":"gpt-4o-mini"}]
}
Input format 
{
  "intent": "...",
  "agents": [
    {"role":"...","policy":"...","provider":"..."}
  ],
  "alignment_mode": "..."
}
Output structure 
{
  "agents":[{"agent_name":"...","decision":"...","output":"...","policy":"...","status":"..."}],
  "summary":{"alignment":"...","conflict":"...","explanation":"..."},
  "signals":["..."]
}
Examples
Customer Support
Request:
{ "intent":"Help a user resolve an issue with a delayed order.", "template":"customer_support_system" }

Response:
{ "agents":[...], "summary":{...}, "signals":[...] }
Financial Decision
Request:
{ "intent":"Advise whether to approve a refund.", "template":"financial_decision_system" }

Response:
{ "agents":[...], "summary":{...}, "signals":[...] }
Uncertain Case
Request:
{ "intent":"User asks about a charge with limited info.", "template":"uncertain_case_system" }

Response:
{ "agents":[...], "summary":{...}, "signals":[...] }
Templates
customer_support — resolve support issues safely
refund_review — structured refund decisions
uncertain — handle ambiguity and escalate
evidence — require proof before answering
conflict — resolve policy conflicts
escalation_handling — safe escalation across agents
multi_step_review — parallel multi-agent review
validation_approval — validate then approve
Output fields
decision → what the system chose
summary → overall outcome
signals → key behavior indicators
/* === M37 CONTROL BUTTON INJECT === */