Featured in OpenAI Agents SDK

Observability, Evaluation, Governance & Policy Enforcement for AI Agents

Enforce, not just observe.

★ Star on GitHub

OpenTelemetry-nativeApache 2.0SOC 2 (in progress)LangChainCrewAIOpenAI Agents SDK

Scroll

01 / 06

How many AI agents are running right now?

→Unified Agent Registry

traccia · governance console

$traccia agents list

142 agents · 3 environments

Customer Support · productionHEALTHY

Data Pipeline · stagingDEGRADED

Code Reviewer · devHEALTHY

142 agents synced

02 / 06

Which agent generated the most LLM cost?

→Cost & Token Attribution

traccia · governance console

$traccia costs top --by agent

Total Trace Cost $14.2045

Input Tokens (GPT-4o)124,500

Output Tokens (GPT-4o)32,100

Embedding Tokens (v3)45,000

computed locally at span-end

03 / 06

Did any agent expose customer data?

→PII & Sensitive Data Detection

traccia · governance console

$traccia violations --severity critical

Critical PII Violation Detected

Patient_Name_ExposureTRIGGERED

Open violations5 alerts

Enforced agents5 blocking

masked before export · init(redact_pii=True)

04 / 06

Did any agent use a restricted LLM model?

→Policy Monitoring & Enforcement

traccia · governance console

$traccia policies status

Tool Call Limits · AgentSoft Block

Restricted Models · WorkspaceTriggered

Cost Guard · OrgHard Block

Compliance Score98.5% Passing

hard blocks stop agents mid-execution

05 / 06

Is this prompt better than production?

→Prompt Registry & Evals

traccia · governance console

$traccia experiments compare --prompt support-reply

dataset refund-cases · 3 scorers

v7 · candidate94% pass

v6 · production81% pass

VerdictImprovement

promote v7 → production · evidence attached

06 / 06

Can you prove compliance?

→Compliance Evidence Built In

traccia · governance console

$traccia evidence export --compliance

✓Art. 12 Governance— Event spans & retention

✓Art. 14 Human Review— Review queue integrated

✓Art. 50 Disclosure— disclosure() via SDK

✓HIPAA Controls— PHI inventory & labeled exports

evidence pack: export-ready

READY

$ traccia.init()

one init call — trace, cost, and govern every agent

traccia · governance console

$traccia.init()

[traccia] registry connected · traces exporting

online — every agent, every framework

traccia · governance console

$traccia agents list

142 agents · 3 environments

Customer Support · productionHEALTHY

Data Pipeline · stagingDEGRADED

Code Reviewer · devHEALTHY

142 agents synced

Unified Visibility

One Dashboard. Every Agent.

Stop switching between LangSmith, custom Grafana dashboards, and spreadsheets. Traccia gives you a single pane of glass for all your AI agents, regardless of framework.

Real-time Health Scores

Monitor agent performance with instant visibility into errors, latency, and throughput.

Cost Attribution

Know exactly which agents and tasks are driving your LLM spend.

Policy Violation Alerts

Instant notifications when agents breach governance rules or spending limits.

Prompt Registry & Evals

Version prompts, grade with datasets and scorers, and promote with experiment evidence.

Compliance Evidence

Export audit-ready packs for EU AI Act and HIPAA Controls from the Governance Hub.

app.traccia.ai

System Healthy

Dashboard

Active Agents5

+1 vs yesterday

Traces / Min42

5% trend

Errors (24h)

142

12% vs prev

Policy Violations23

0% flat trend

Token Spend$1,250.45

8% vs prev

Agents

System inventory and status

Agent Name	Status	Owner	Env	Executions	Cost
Web Researcher	HEALTHY	J Avam...	production	248	$220.00
Zendesk Copilot	DEGRADED	J Avatar...	staging	27	$12.00
Stripe Risk Evaluator	CRITICAL	MA Kivin	dev	18	$100.00
Data Scientist GPT	HEALTHY	J Avam...	production	30	$13.00
Shopify Support Bot	HEALTHY	I Avatar	staging	20	$7.00

View all agents

Cost Snapshot - Last 24 hours

$1,250.45

Top Spenders

Web Researcher$1,250.45

Zendesk Copilot$12.00

Top Models

GPT-4$7.00

Claude 3.5 Sonnet$0.00

System Behavior

Retry Rate

9.2%

HIGH

Error Rate

3.4%

NORMAL

Active Agents

5 Monitored

Cost Saved Today

$127.40

The Control Plane

Stoptogglingbetweentools.

Everything you need to orchestrate your AI agents with Swiss precision.

Unified Agent Registry

One source of truth for all agents across LangChain, CrewAI, and OpenAI Agents SDK. Track version, health, environment, and ownership — zero code changes.

Agent Registry142 agents

Customer Supportproduction

HEALTHY

Data Pipelinestaging

DEGRADED

Code Reviewerdev

HEALTHY

System Status142 agents synced

End-to-End Tracing

Full OpenTelemetry-native visibility into every LLM call, tool use, and agent decision. Supports LangChain, CrewAI, OpenAI Agents SDK, AutoGen, and LlamaIndex.

Execution TimelineMASKED

0ms1.25s2.50s

user_input_processing

knowledge_base_search

llm_generation

response_masking

Total Cost

$0.0051

LLM Calls

3 calls

Trace ID72d76daf2693...

Policy Monitoring & Enforcement

3-tier guardrail detection across explicit annotations, provider-native signals (OpenAI, Anthropic, Google), and heuristics. Hard blocks stop agents mid-execution — not after the fact.

Active Policies3 RULES

Tool Call LimitsAgent Scope

INFOSoft Block

Restricted ModelsWorkspace Scope

WARNINGSoft Block

Cost GuardOrg Scope

CRITICALHard Block

Open Violations

5 alerts

Enforced Agents

5 blocking

Compliance Score98.5% Passing

Cost & Token Attribution

Accurate cost across 2,500+ models. Computed locally at span-end, independent of sampling. Token and cost totals stay 100% accurate — even at 10% trace sampling.

Total Trace CostLive

$14.2045

computed locally at span-end

Input Tokens (GPT-4o)124,500

Output Tokens (GPT-4o)32,100

Embedding Tokens (v3)45,000

PII & Sensitive Data Detection

Flags PII exposure — patient names, credentials, user data — in agent traces before they reach your logs. Annotate violations, enforce hard blocks, stay compliant automatically.

ViolationsCritical

5 alerts

Open violations

5 blocking

Enforced agents

Critical PII Violation Detected

Patient_Name_ExposureTRIGGERED

Prompt Registry & Playground

Named prompts with immutable versions and a protected production label. Compare wording and models side by side with latency, tokens, and cost, then load the live version from the SDK so every trace links back to what shipped.

support-replyCHAT · 7 VERSIONS

v7Tighten refund policy wording

production

v6Add {{account_tier}} variable

—

v5Soften greeting; shorten system

—

Playground · side by side

3 panels · latency · tokens · costCOMPARE

load_prompt · productionv7 · linked in trace

Datasets, Scorers & Experiments

Curate test cases, grade with built-in checks, LLM-as-judge, or custom scorers, and save immutable experiment reports. Compare baseline vs candidate, then promote with evidence attached.

Experiment CompareIMPROVEMENT

v7 · candidaterefund-cases · 48 rows

94% PASS

v6 · baselinesame dataset · 3 scorers

81% PASS

Built-in · LLM-as-judge · customscorers snapshotted

Promote to productionevidence attached

Compliance Evidence Built In

Article-mapped EU AI Act evidence out of the box — Art. 12 GovernanceEvent spans, Art. 14 human review, Art. 50 disclosure() in the SDK. Opt into HIPAA Controls for PHI-capable agent inventory, safeguard drafts, and labeled exports. Export evidence packs in one click.

Compliance ChecklistEU · HIPAA

Art. 12 Governance

Event spans & retention

Art. 14 Human Review

Review queue integrated

Art. 50 Disclosure

disclosure() via SDK

HIPAA Controls

PHI inventory & labeled exports

Evidence Packexport-ready

OpenTelemetry-Native

OneSDK. Every Framework.

Drop-in tracing for LangChain, CrewAI, OpenAI Agents SDK, AutoGen, and LlamaIndex.

Works with any LLM: OpenAI & Anthropic include automatic token & cost tracking; any HTTP-based provider is traced via OpenTelemetry, with deeper coverage added continuously.

Key Differentiators

Whyteamschoose Traccia

The only tool that enforces, not just observes

Others alert you after the fact. Traccia’s policy engine hard-blocks agents mid-execution — stopping runaway costs, infinite loops, and PII leaks before they hit production.

Guardrail detection with zero runtime overhead

A passive span processor inspects every trace as it ends. Three detection tiers — explicit, provider-native, heuristic — with no changes to your agent code and zero added latency.

Cost totals that don’t lie at 10% sampling

Sampling cuts trace volume but your costs don’t scale with it. Traccia emits OTEL metrics for every LLM call independently — token and cost totals stay 100% accurate regardless of sample rate.

One registry, not locked to one SDK

LangSmith works best if you use LangChain. Traccia is OpenTelemetry-native — works across every framework today and every one you adopt next quarter. Switch in one afternoon with pip install traccia.

Enterprise Ready

Security that enterprises trust

Built for teams that need observability, auditability, and governance at scale.

Enterprise Security

SOC 2 Type II certification in progress. Security controls designed for enterprise procurement.

PII redaction (SDK)

Optional pattern-based masking before export via init(redact_pii=True). ML PII detection planned.

RBAC & Access Control

Granular role-based access control today. OIDC/SAML SSO on the enterprise roadmap.

Start building with Traccia today.

Choose your path to start tracing and monitoring your AI agents

Use SDK Locally

Open Source

Export traces to your own observability stack. Works with Jaeger, Grafana Tempo, Zipkin, SigNoz, and any OpenTelemetry-compatible backend.

No account required
OTLP standard export
Full SDK control

Use Traccia Platform

Full Power

Access the complete Traccia dashboard with agent-first governance, cost attribution, policy guardrails, and team collaboration.

Policy guardrails
Cost attribution
Prompt registry & evals