A vectorless RAG engine that builds a tree index of long PDFs and lets an LLM reason its way to the right section, with page-level citations.
Traccia
Summary
An OpenTelemetry-native control plane for AI agents that enforces policy mid-run instead of paging you after the incident, with cost attribution and EU AI Act evidence built in.
Screenshots
Description
Traccia is a control plane for teams running AI agents in production. The pitch it leads with is "enforce, not just observe": where a tracing tool tells you afterwards that an agent leaked PII or burned a month of budget in an afternoon, Traccia can stop the run while it is happening.
Observability
Instrumentation is OpenTelemetry-native, so traces cover LLM calls, tool invocations and agent decisions end to end and can be exported to whatever backend you already run. The dashboard tracks agents across environments with health scores, error rates and latency, and every span is inspectable down to the decision that produced it.
Enforcement
Policies are evaluated in-line with three tiers of detection — explicit rules, provider-native signals and heuristics — and can hard-block an agent mid-execution on a cost overrun or a policy violation. Sensitive-data detection flags PII exposure before logs leave the system, which is the difference between a control and an audit finding.
Cost and compliance
Token and cost attribution is computed locally at span-end rather than derived from sampled traces, so the numbers hold up when sampling is aggressive. A prompt registry versions prompts and runs A/B evaluations against them, and compliance evidence exports are shaped for EU AI Act and HIPAA reporting.
Coverage
One SDK covers LangChain, CrewAI, the OpenAI Agents SDK, AutoGen and LlamaIndex, against 2,500+ models across OpenAI, Anthropic and generic HTTP providers. The local SDK is open source under Apache 2.0 and exports to any OpenTelemetry backend — Jaeger, Grafana Tempo, Zipkin — so the instrumentation layer is usable without the hosted platform.
Pricing
Hobby is free with 50K events and 7-day retention. Observe is $99/month for 500K events, $12 per additional 100K, 30-day retention. Govern is $299/month for 2M events, $8 per 100K, 90-day retention. Scale is $799/month for 10M events, $5 per 100K, one-year retention. Enterprise is custom with ten-year retention. An event is any trace, span or LLM call Traccia observes, and every paid plan carries unlimited users. The team is based in India and reports SOC 2 Type II certification in progress.
Reviews
Similar App Suggestions
Agentic data platform that turns business questions into defensible, auditable narratives — with sources, assumptions and lineage attached.
Reads your production chat and voice agent conversations to surface the silent failures, frustration loops and policy breaches that offline evals never catch.
Web data extraction APIs for AI agents — 75+ ready-made scrapers for LinkedIn, Amazon, Google Maps, Reddit and X, priced per result.
Pipeline, encrypted Iceberg lake and query engine in one, with an MCP server that lets Claude and Codex query your data directly.