Fiddler AI

Fiddler AI is an enterprise observability, evaluation and guardrails platform for machine learning models, LLM applications and multi-step AI agents in production. It gives data science, engineering, trust and safety, and security teams one place to trace agent behaviour, catch drift and hallucination, enforce policy inline, and produce audit-ready governance records for regulated deployments.

Fiddler AI, founded in 2018 by Krishna Gade and Amit Paka and headquartered in Palo Alto, California, builds what it now calls an AI control plane: a single platform for evaluating, monitoring, governing and guarding machine learning models, LLM applications and multi-step AI agents in production. The platform began as an explainable-AI and ML observability tool — drift detection, performance monitoring, and root-cause analysis down to the exact data slice causing an alert — and has since extended into agentic systems, adding distributed tracing across agent workflows that span models, tools and external APIs. Its guardrails tier runs on Fiddler's own fine-tuned Centor Models, which score requests and responses inline for hallucination, toxicity, PII and PHI exposure, prompt injection and jailbreak attempts; the vendor advertises sub-80-millisecond enforcement latency and in-environment processing so payloads never leave the customer's boundary. Instrumentation is OpenTelemetry-based, with native Python and JavaScript SDKs plus dedicated packages for LangGraph, LangChain, Google ADK and Strands Agents, and connectors into Snowflake, BigQuery, Databricks, MLflow, Kafka, Amazon S3, Airflow, Datadog and PagerDuty. Fiddler sells to regulated and public-sector buyers — Nielsen, MasterCard, Ally, Integral Ad Science, American Family Insurance, CSAA, Thumbtack and the U.S. Navy are named publicly — and offers SaaS, VPC, AWS GovCloud and on-premises deployment alongside an Amazon SageMaker Partner AI App listing. The company raised a $30M Series C led by RPS Ventures in January 2026, taking total funding to $100M, and reports more than 4x revenue growth over the preceding 18 months.

Category

Governance & Security

Pricing

Freemium, Usage-based, Contact for pricing

Target Market

CIOs, CTOs, Data Scientists, Enterprise Developers, CISOs, MLOps Engineers

Deployment

Cloud-first, Hybrid, Self-hosted

Founded

2018

Headquarters

Palo Alto, United States

• ✓
  Agentic observability

  End-to-end distributed tracing across multi-step agent workflows spanning models, tools and external APIs, so a failure can be traced to the hop that caused it.

• ✓
  Inline guardrails on Centor Models

  Fiddler's own fine-tuned scoring models enforce policy on the live request/response path in under 80ms, blocking jailbreaks, PII/PHI and injection attempts.

• ✓
  Continuous evaluations

  Runs the same evaluators from pre-deployment testing through production using built-in metrics, custom evaluators and bring-your-own LLM-as-a-judge.

• ✓
  ML observability with explainability

  Drift detection and performance monitoring paired with explainability, which reviewers say is what lets them trace an alert to the exact data slice.

• ✓
  OpenTelemetry-native instrumentation

  Python and JavaScript SDKs plus OTLP support and framework packages for LangGraph, LangChain, Google ADK and Strands, avoiding proprietary lock-in on telemetry.

• ✓
  Flexible enterprise deployment

  Runs as SaaS, inside a customer VPC, on AWS GovCloud or fully on-premises, so regulated data never has to leave the customer boundary.

• ✓
  Governance and audit records

  Centralised policy definition with audit-ready records of what was enforced and when, which is what risk and compliance teams actually ask for.

• •
  Putting a customer-facing agent into production safely

  Trace every agent step, enforce guardrails inline on the response path, and keep a record of what was blocked for later audit.

• •
  Catching silent model degradation

  Drift detection alerts on distribution shift and explainability narrows it to the feature and data slice responsible, cutting triage from days to hours.

• •
  Meeting AI governance and regulatory obligations

  Produce centralised, auditable evidence of model behaviour and policy enforcement for internal risk committees, external auditors or a regulator.

• •
  Preventing data exfiltration through LLM prompts

  Detect and redact PII, PHI and secrets in requests and responses in-environment, so sensitive payloads never reach a third-party model provider.

• •
  Consolidating a split ML and GenAI observability stack

  Run classic model monitoring and LLM/agent tracing on one platform instead of paying for and correlating two separate toolchains.

Unusually for this category Fiddler publishes a rate: the Developer tier is metered at $0.002 per trace, and the guardrails-only Free tier costs nothing, which makes a proof of concept cheap to model in advance. Everything an enterprise actually needs is gated above it — VPC and on-premises deployment, enterprise-grade guardrails, scaled infrastructure and named support are Enterprise-only and quoted, with no list price published. Budget on trace volume, not seats, and expect the deployment model rather than the feature list to be what pushes you into a sales conversation.

This page was written from 7 sources, 4 on domains other than fiddler.ai.

1. 1.fiddler.ai — fiddler.aivendor
2. 2.fiddler.ai — pricingvendor
3. 3.fiddler.ai — securityvendor
4. 4.docs.fiddler.ai — llms.txt
5. 5.peerspot.com — fiddler ai reviews
6. 6.pulse2.com — fiddler 30 million series c
7. 7.hn.algolia.com — search