Fiddler AI
Also known as: Fiddler
Fiddler AI is an AI control plane for enterprise agents: hierarchical agent tracing, evaluations, and real-time guardrails powered by in-environment Centor Models, with SaaS, VPC, on-premises and air-gapped deployment.
Fiddler AI calls itself the AI control plane for enterprise agents, one platform to observe, evaluate and enforce policy across first party, third party and coding agents as well as traditional ML models. It traces agents hierarchically from application to session, agent, trace and span through native SDKs for LangGraph, LangChain, Google ADK and Strands, OpenTelemetry, S3 ingestion and gateways such as Kong, AgentGateway and LiteLLM, and runs evaluations with its Evals SDK, pre-built and custom LLM-as-a-judge evaluators and more than 100 metrics.
Its purpose-built Centor Models, formerly Trust Models, run entirely in the customer's environment and power the Fiddler Trust Service and real-time guardrails, which enforce policy inline on an agent's request and response path in under 100 milliseconds against hallucinations, toxicity, PII and PHI leakage, prompt injection and jailbreaks, with inline redaction of secrets for coding agents. A Fiddler MCP server lets AI assistants query applications, traces and evaluator results, and governance features align with the NIST AI RMF and ISO/IEC 42001, GDPR and HIPAA.
Fiddler states SOC 2 Type II and HIPAA compliance, with SSO and role based access control, and deploys as SaaS, in a VPC, on premises, in AWS GovCloud, air gapped, or inside Amazon SageMaker. A Free plan covers the guardrails, a Developer plan is usage-based at $0.002 per trace, and Enterprise adds flexible deployment and dedicated support. Customers include Nielsen, the U.S. Navy and Integral Ad Science.
Vendor details
Canonical URL
https://www.fiddler.ai
Category
Agent infrastructure
Subcategory
Observability and security
Funding status
Fiddler is independent. It was founded in 2018 in Palo Alto and raised a $30 million Series C in January 2026 led by RPS Ventures, bringing total funding above $60 million. Its Fortune 500 customers include Nielsen.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Fiddler integrates into existing MLOps pipelines and through the gateway a team already runs, monitoring agents, LLM applications, and predictive ML across tabular, text, and image data. It offers more than a hundred out of the box and custom metrics, an SDK, and an API, with cloud and VPC deployment.
In practice
Your agents make autonomous decisions and leadership asks whether you can control them. Fiddler gives unified observability from application to span, deep diagnostics for agent failures, and governance across the lifecycle.
A regulated deployment needs real time protection on outputs. The Fiddler Trust Service applies fast guardrails for moderation and harmful exposure, with audit trails for GDPR and HIPAA obligations.
You run predictive models and LLM agents and want one pane. Fiddler monitors both across tabular, text, and image data with over a hundred metrics including drift, hallucination, and PII detection.
Sources & related URLs
Agentic Index coverage score
8.5 / 14 capabilities · 61%
| Integrations & Tool Calling | Partial |
|---|---|
|
Fiddler takes in traces from agent frameworks, OpenTelemetry, S3 and AI gateways (Kong, AgentGateway, LiteLLM), reads model data from Snowflake, S3, SageMaker, Databricks and MLflow, and sends alerts to Datadog and PagerDuty. These move telemetry in and alerts out, and there are no connectors that let an agent take authenticated actions in outside systems. SourceFiddler, docs.fiddler.ai llms.txtread 2026-09-21 |
|
| Workflow Orchestration | Not documented |
|
There are no workflows that sequence, branch or retry an agent's steps. Fiddler observes, evaluates and guards agents that run elsewhere. SourceFiddler, fiddler.ai llms.txtread 2026-09-21 |
|
| Knowledge Grounding & RAG | Not documented |
|
There is no retrieval structure over the customer's knowledge that grounds an agent's answers. Fiddler evaluates RAG applications built elsewhere, including checks for faithfulness and hallucination. SourceFiddler, docs.fiddler.ai llms.txtread 2026-09-21 |
|
| Human Oversight & Guardrails | Full |
|
Guardrails enforce policy inline on an agent's request and response path in under 100 milliseconds, detecting hallucinations, toxicity, PII and PHI leakage, prompt injection and jailbreaks. For coding agents they redact PII, PHI and secrets inline through the customer's existing LLM gateway, and enforceable policy is one of the platform's five pillars. SourceFiddler, fiddler.ai llms.txt and homepage, and docs.fiddler.ai llms.txt (AI gateway guardrail enforcement)read 2026-09-21 |
|
| Security, Identity & Governance | Full |
|
The SOC 2 Type II report covers security, confidentiality and availability, is audited every year and is available under NDA, and Fiddler states HIPAA compliance. Customers sign in through SSO providers and grant granular permissions with role based access control, data is encrypted with AES-256 at rest and TLS 1.2 or higher in transit, and customer data is kept and deleted under a retention policy. SourceFiddler, fiddler.ai/securityread 2026-09-21 |
|
| Observability & Auditability | Full |
|
Visibility runs from application to session, agent, trace and span. There are native SDKs for LangGraph, LangChain, Google ADK and Strands, OpenTelemetry ingestion, tracing of LLM calls and tool use for coding agents through Claude Code and existing gateways, alerts routed to Datadog and PagerDuty, and customer data kept under a stated retention policy. SourceFiddler, fiddler.ai llms.txt, docs.fiddler.ai llms.txt and fiddler.ai/securityread 2026-09-21 |
|
| Memory & State Persistence | Not documented |
|
There is no session, conversation or long term memory that an agent reads back as context. Fiddler records sessions and traces for observability. SourceFiddler, fiddler.ai llms.txtread 2026-09-21 |
|
| Deployment & Data Residency | Full |
|
Deployment runs as SaaS, in the customer's VPC, on premises, in AWS GovCloud or air-gapped, and as a Partner AI App inside the customer's Amazon SageMaker, with Fiddler's Centor evaluation models running entirely in the customer's environment. SourceFiddler, fiddler.ai homepage and llms.txt, and docs.fiddler.ai llms.txtread 2026-09-21 |
|
| Prebuilt Agents, Templates & Packs | Not documented |
|
There are no ready made agents, workflows or templates a customer adopts and runs. Fiddler ships prebuilt evaluators, a metrics library and cookbooks for common uses. SourceFiddler, docs.fiddler.ai llms.txt and fiddler.ai llms.txtread 2026-09-21 |
|
| Triggers & Channel Coverage | Full |
|
Guardrails evaluate every agent request and response inline as it passes, continuous evaluations score production traces, and alert rules fire on monitored metrics, so Fiddler's evaluation and enforcement start from each event with no person initiating them. SourceFiddler, fiddler.ai llms.txt and homepage, and docs.fiddler.ai llms.txtread 2026-09-21 |
|
| Model Flexibility & Routing | Full |
|
Custom prompts and a customer choice of judge model drive Fiddler's LLM-as-a-judge evaluators, beside its own Centor Models that run in the customer's environment, so the customer chooses the model behind Fiddler's own evaluation features. Fiddler does not publish which judge models are supported. SourceFiddler, docs.fiddler.ai llms.txt (Advanced Prompt Specs) and fiddler.ai llms.txtread 2026-09-21 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
Native SDKs include Fiddler's OpenTelemetry SDK plus LangGraph, LangChain, Google ADK, Strands and Evals SDKs, alongside a Python client for alerts and other resources and a Fiddler MCP server that lets assistants that speak MCP query applications, traces and evaluator results. SourceFiddler, docs.fiddler.ai llms.txtread 2026-09-21 |
|
| Testing, Debugging & Optimization | Full |
|
The Fiddler Evals SDK runs experiments on LLM applications with pre-built evaluators and custom metrics, datasets with golden labels and side-by-side comparison, custom LLM-as-a-judge evaluators, and continuous evaluation in production over more than 100 metrics, scored by Fiddler's Centor Models in the customer's environment. SourceFiddler, docs.fiddler.ai llms.txt and fiddler.ai llms.txtread 2026-09-21 |
|
| Browser & Computer Use | Not documented |
|
Agents get no browser, desktop or computer control, since Fiddler observes and guards agents. SourceFiddler, fiddler.ai llms.txtread 2026-09-21 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Product updates
Evals datasets in Fiddler can now be edited through the REST API in Release 26.25, with every add, update or delete saved as a new numbered version that can still be read later. A write based on an out of date version is rejected and nothing changes. The Fiddler MCP server can now make the same edits in a single call and list a dataset's versions.
Bears on: Observability / auditability
View sourceRelease 26.24 adds secret detection to Fiddler's Sensitive Information guardrail, flagging API keys, tokens, credentials and high entropy strings across more than 35 known secret formats as a new entity type in the PII categories. A validation layer also checks URLs, IP addresses, emails, SSNs and credit card numbers against format rules to cut false positives.
Bears on: Security / enterprise
View sourceFiddler shipped a Fiddler MCP Server with OAuth 2.0 authentication for clients such as Claude Code, along with trace routing across multiple applications, in releases 26.15 to 26.19.
Bears on: MCP / tool calling / API
View sourcePricing
Free plan (guardrails) · Developer $0.002 per trace · Enterprise custom
Usage-based on the Developer plan, at $0.002 per trace; Enterprise quoted.
Included quota
Pricing is tiered as Lite, Business, and Premium, all requiring a sales conversation, and billed on consumption such as data ingested, number of models, and metrics.
What is public
Fiddler publishes a Free plan with real-time guardrails, a usage-based Developer plan at $0.002 per trace, and an Enterprise plan; the plan details come from the homepage FAQ.
Billing mechanics
The Developer plan is usage-based at $0.002 per trace. Enterprise is set through sales, with SaaS, VPC or on-premises deployment and dedicated support.
Cost watchouts
Developer billing is per trace, so busy multi-step agents add up quickly; deployment in a VPC or on premises sits on Enterprise.
Variable cost rationale
The Developer plan bills per trace, so cost grows directly with agent traffic.
Overage / add-ons
Consumption based; costs rise with data ingested, models, and metrics under the contracted tier.
Sales call required
Mixed (some tiers require a call)
Free / trial
Free plan with real-time guardrails
Lowest paid plan
Developer, usage-based at $0.002 per trace
Commercial notes
Fiddler announced a $30 million Series C in 2026. Customers include Nielsen, the U.S. Navy and Integral Ad Science.
Key ambiguities
Plan details come from the homepage FAQ; how the Developer plan is bought is not established, so check the pricing page or ask Fiddler.
Related vendors
- AgentOps — Agent observability and debugging platform: open source SDKs trace…
- AgentPhone — AgentPhone gives the AI agents a team builds their own US or…
- Agno — Python agent framework and AgentOS runtime (formerly Phidata) for…
- AIsa — Resource and payment gateway for AI agents, with one key to 110+…
- Albato — Albato is a no code integration platform that sells an embedded,…
- AlphaBitCore — AI control plane for regulated financial firms: one gateway enforces…
Alternatives to Fiddler AI
The closest documented capability profiles to Fiddler AI among agent infrastructure platforms tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- F5 AI Guardrails9.0 / 14Adds documented Prebuilt Agents, Templates & Packs
- Freeplay8.0 / 14A lighter documented profile than Fiddler AI
- Galileo9.0 / 14Adds documented Prebuilt Agents, Templates & Packs
- Helicone8.0 / 14A lighter documented profile than Fiddler AI
- Opik9.0 / 14Adds documented Prebuilt Agents, Templates & Packs
- Traceloop8.0 / 14A lighter documented profile than Fiddler AI
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded