Traccia
AIOpenTelemetry-native observability for AI agents that also enforces policy at runtime, with cost attribution and EU AI Act evidence export.
Traccia
OpenTelemetry-native observability for AI agents that also enforces policy at runtime, with cost attribution and EU AI Act evidence export.
About Traccia
Traccia is a control plane for AI agents whose pitch is enforcement rather than dashboards: it observes every agent execution and can also stop one mid-run. A single traccia.init() call registers the agent and starts exporting OpenTelemetry-native traces covering every LLM call, tool use and agent decision across LangChain, CrewAI, the OpenAI Agents SDK, AutoGen and LlamaIndex, building a unified registry that tracks version, health, environment and ownership with no code changes per agent. On top of the trace stream it computes cost and token attribution locally at span end, so you can see which agent and which task drove LLM spend, detects and masks PII before export, and runs policies — tool call limits, restricted model lists, cost guards — as soft or hard blocks that halt execution when breached. A prompt registry with datasets and scorers lets teams version prompts, compare a candidate against production, and promote with the experiment evidence attached. The governance hub exports audit-ready evidence packs mapped to EU AI Act articles 12, 14 and 50 and to HIPAA controls. Traccia is Apache 2.0, installs with pip, retains 100% of data with no sampling, and charges by events rather than seats.
Screenshots
Key Features
Use Cases
Frequently asked questions about Traccia
What is Traccia?
OpenTelemetry-native observability for AI agents that also enforces policy at runtime, with cost attribution and EU AI Act evidence export.
How does Traccia work?
Traccia works by combining Runtime Policy Enforcement: Tool call limits, restricted model lists and cost guards run as soft or hard blocks that stop an agent mid-execution, not just alert after the fact., Unified Agent Registry: One source of truth listing every agent across frameworks and environments with version, health, ownership and sync status, with zero code changes per agent., OpenTelemetry-Native Tracing: End-to-end spans over every LLM call, tool use and agent decision, exportable to the OTel tooling you already run., Cost and Token Attribution: Input, output and embedding tokens priced and computed locally at span end, rolled up per agent and per task to show top spenders., PII and Sensitive Data Detection: Detects exposure of customer data in agent traffic and masks it before export, with init(redact_pii=True) as the switch. to help users with Agent Fleet Visibility: Replace a mix of LangSmith, custom Grafana dashboards and spreadsheets with one view of every agent regardless of framework., LLM Spend Control: Find which agents and tasks are driving cost, then set hard spend caps that stop runaway loops before the invoice does., Regulated Deployment: Produce audit-ready EU AI Act or HIPAA evidence packs on demand instead of reconstructing them under deadline., Data Leak Prevention: Catch and mask PII in agent traces before it leaves the system, with critical violations surfaced as alerts., Prompt Change Management: Prove a candidate prompt beats production on a scored dataset before promoting it, with the experiment kept as evidence..
What are the main features of Traccia?
Key features include Runtime Policy Enforcement: Tool call limits, restricted model lists and cost guards run as soft or hard blocks that stop an agent mid-execution, not just alert after the fact., Unified Agent Registry: One source of truth listing every agent across frameworks and environments with version, health, ownership and sync status, with zero code changes per agent., OpenTelemetry-Native Tracing: End-to-end spans over every LLM call, tool use and agent decision, exportable to the OTel tooling you already run., Cost and Token Attribution: Input, output and embedding tokens priced and computed locally at span end, rolled up per agent and per task to show top spenders., PII and Sensitive Data Detection: Detects exposure of customer data in agent traffic and masks it before export, with init(redact_pii=True) as the switch..
Who is Traccia for?
Traccia is useful for anyone interested in Agent Fleet Visibility: Replace a mix of LangSmith, custom Grafana dashboards and spreadsheets with one view of every agent regardless of framework., LLM Spend Control: Find which agents and tasks are driving cost, then set hard spend caps that stop runaway loops before the invoice does., Regulated Deployment: Produce audit-ready EU AI Act or HIPAA evidence packs on demand instead of reconstructing them under deadline., Data Leak Prevention: Catch and mask PII in agent traces before it leaves the system, with critical violations surfaced as alerts., Prompt Change Management: Prove a candidate prompt beats production on a scored dataset before promoting it, with the experiment kept as evidence..
How much does Traccia cost?
Traccia offers a free tier with paid plans for advanced features.
How do I get started with Traccia?
Visit https://traccia.ai to sign up and explore Traccia.
Explore more AI Ai Tools tools
Browse all Ai Tools tools →Compare Traccia: vs Mise · vs BiBimba · vs VoiceCap · vs NM Signals