Trace, debug, and control every AI agent decision.

AISquare captures the reasoning behind every agent decision. Debug faster, catch regressions before they ship, and build AI systems that are explainable, reliable, and production-ready.

→Connect with any agent stack
→See why. Not just what happened.
→Debug and fix. No redeploy cycle.
→Govern and scale AI Agents.

Read the docs Contribute to open source

reasoning.trace

Live

run #4821agent ResearchAssistant1.2s5 nodes

1 assumption

Execution Graph

consumed

exec

produced

datatext/plaind2e6

agentResearchAssist...

llmgpt-4o

toolwiki_search

dataapplication/json69b5

Info

XTrace

RML

Policies

Extraction Conf.

87%

Claims

2

1.Q3 revenue projected at $4.2M based on current pipeline

2.Growth driven by enterprise segment expansion

Assumptions

1

YoY growth rate 18% applied

Not found in retrieved context. Applied from prior run.

Evidence Attribution

finance_db

92%

quarterly_v2.pdf

78%

View Reasoning

Edit Policies

Review needed

Agent debugging is still guesswork.

You can trace every run. You still struggle to debug it.

run_a→tool_call("web_search")

run_b→tool_call("calculator")

Same input. Different output. No idea why.

Your traces show both runs. They don't explain why your agent chose different paths.

0.91·0.91·0.91·0.62↓ DROPPED

changed_by: ???

Something changed. Your traces can't say what.

Behavior shifted after a deployment. Pinpointing the cause is hours in the traces. Sometimes a full day.

fix(prompt): attempt_8 → redeploystill wrong. try again.// ...repeat until something sticks

Found the bug. Now the real work starts.

Fixing one bad decision means cycling through traces, prompts, and redeploys until something sticks.

HOW IT WORKS

Add trust to your AI development workflow

Connect in minutes. Trace every decision. Fix behavior before it ships.

01

STEP 01

Connect to your stack in minutes

Install the SDK or connect through the proxy. AISquare starts capturing agent runs without forcing your team into a new workflow.

Quick Setup

Set up tracing in your AI pipeline in minutes

1Generate API Key

Generate API Key for This Studio

2Install the SDK

pip install aisquare[explainability]

3Quick Start

# Auto-configure from environment

import aisquare.explainability as sdk

sdk.init_from_env(service_name="my-agent")

# AgentRunTracer wraps your agent

with AgentRunTracer(agent_name="MyAgent") as run:

run.set_output("Agent response")

Claude Code ProxyRunning

Open Policy AgentConnected

02

STEP 02

Trace every decision as it happens

Every agent run produces a complete execution graph. See inputs, tool calls, reasoning steps, and outputs in one place so your team can understand how each run actually happened.

GraphOverview2 tool calls⚠ 1 POLICY VIOLATION

⚡ TRACES

1 VIOLATION2m ago

Create 5 Pyt…

COMPLETED5m ago

check the files

COMPLETED12m ago

summarize repo

COMPLETED1h ago

refactor utils

ⓘ INFO

DURATION

23.6s

COST

$0.21

TOKENS

67,918

MODEL

gpt-4o

TOOLS

2 calls

03

STEP 03

Catch what exactly broke across runs

Compare runs, click the flagged node, and pinpoint why behavior shifted after a prompt tweak, tool update, context change, or release. Everything you need to understand the issue in one view.

reasoning.trace · LLM node selected⚠ 2 policy violations

LLMgpt-4o · call_7f2a

Confidence61%

Tokens2,847 / 312

2 policy violations on this node

api.security.llm_output

possible injection pattern detected

FLAGGED

api.security.owasp_top10

possible injection pattern detected

FLAGGED

Assumption not grounded

"YoY growth rate 18%" not found in context. Carried from prior run.

04

STEP 04

Fix behavior where it breaks

Apply targeted controls, policies, and interventions instead of bouncing between traces, prompts, code, and redeploys. No prompt trial-and-error.

reasoning.trace · Policies⚠ 7 policy violations

InfoTraceRMLPoliciesAttestations

7 files with policy violations

database.py1 ISSUEWRITE

/Users/mohit/.../database.py

⚠ WARNING

Secret/Credential Leak Detection

Potential secret in file write content

💡 How to Fix🔑 Auto fix

user_repo.py1 ISSUEWRITE

product_repo.py1 ISSUEWRITE

✱AISquare Fixed Files1

✱database.py1 FIXED

/Users/mohit/.../database.py

WHAT WAS WRONG

⚠ WARNING

Secret/Credential Leak Detection

Potential secret in file write content

HOW IT WAS FIXED — DIFF

@@ -1,11 +1,13 @@

import psycopg2

+ import os

def get_connection():

- password="secret123"

+ password=os.getenv("DB_PASSWORD")

+ # [AISquare Fix] owasp-a06-secrets

05

STEP 05

Fix once. Every future run covered.

Turn the fix into persistent policy and feedback so the same failure does not keep coming back.

reasoning.trace · AI System Overview✔ Governing 39 runs

AI System Overview

Updated 03:32 PM

7d14d30d

TOTAL RUNS

39

7d window

SUCCESS RATE

84.6%

33 successful runs

POLICY VIOLATIONS

11

84.6% compliance

AVG DURATION

35.3s

avg cost $0.271

POLICY HEALTH

Compliance rate84.6%

BY SEVERITY

Critical

3

Warning

8

TOP VIOLATED POLICIES

#1

api.security.owasp_top10

ERROR9

#2

api.security.llm_output

ERROR2

#3

api.security.llm_input

ERROR1

7 days · 39 runs · 11 violations$10.29

Start integrating AISquare into your AI stack

Connect AISquare to the agent frameworks, model providers, and runtimes your team already uses.

Agno

LangChain

LlamaIndex

CrewAI

OpenAI

Claude

Gemini

Mistral

OpenClaw

Docs

Guides, quickstarts, and concepts to get you building fast.

Read docs

API Reference

Explore endpoints, schemas, and integration patterns for AISquare systems.

Explore API

SDKs

Drop-in Python SDK to add explainability to any AI agent or pipeline.

Get SDK

Open Source

A curated list of AISquare projects, integrations, and community resources.

View repos

Stay up to date

Events, insights, and release notes. Everything happening at AISquare, in one place.

Build with AISquare

Start building, contribute in public, and get help when you need it.

Open Source

Explore starter repos, integrations, and example projects or contribute to the ecosystem as it grows.

AISquare-Studio-QA
AI-powered GitHub Action that converts natural language test descriptions into Playwright tests
django-ais
Django-native orchestration framework for agentic workflows with DB-backed job management and real-time event streaming
aisquare-integrations
Adapters for LangChain, CrewAI, AutoGen
Coming Soon

Star on GitHub

Join the community

500+ developers building with AISquare. Ask questions, share what you're building, and stay ahead of what's shipping next.

500+

Developers

24h

Avg response time

Join Discord

Bring trust, traceability, and control into your AI workflow

Get a complete picture of every decision your agent makes.

Get API Keys Read the docs

AISquare is the audit, trust, and governance layer for AI systems.

Trace, debug, and control every AI agent decision.

Agent debugging is still guesswork.

Same input. Different output. No idea why.

Something changed. Your traces can't say what.

Found the bug. Now the real work starts.

Add trust to your AI development workflow

Connect to your stack in minutes

Trace every decision as it happens

Catch what exactly broke across runs

Fix behavior where it breaks

Fix once. Every future run covered.

Start integrating AISquare into your AI stack

Docs

API Reference

SDKs

Open Source

Stay up to date

Events

Latest Insights

Changelog

Build with AISquare

Open Source

Join the community

Bring trust, traceability, and control into your AI workflow