These are the two closest products in this market: both trace agents, score production traffic, offer small judge models and add guardrails at run time. Galileo, now Splunk Agent Observability, checks tool inputs and outputs during a run through Agent Control; Fiddler is independent, covers classic ML models too and sells a self-serve per-trace plan. Whether you buy through Splunk often decides it.[1][6]
Both pair evaluation with run-time guardrails and both can run on-premises; the difference is ownership and where the guardrails sit. Galileo is now sold as Splunk Agent Observability after Cisco completed its acquisition in May 2026, and its Agent Control checks model and tool inputs and outputs during a run. Fiddler is independent, also monitors classic ML models, and scores content in real time for your application or an AI gateway to block.
| Decision factor | Fiddler | Galileo |
|---|---|---|
| Ownership | Independent; $30M Series C in January 2026 | Acquired by Cisco (completed May 2026); sold as Splunk Agent Observability since 7 August 2026 |
| Run-time checks | Partial: real-time content scores; inline blocking through a third-party AI gateway | Yes: Agent Control checks LLM and tool inputs and outputs during execution |
| Judge models | Centor Models and a hosted Ministral 3 8B judge | Luna-2 small models; five preset Luna evaluators in the Splunk-hosted edition |
| Human review | Span annotations, in public preview | Annotations and annotation queues (generally available August 2026) |
| Prompt tooling | None | Prompts and a playground; version control not documented |
| Classic ML monitoring | Yes: drift, performance, explainability, fairness | Not documented |
| Self-hosting | Enterprise: VPC, on-premises or SageMaker in your AWS account | On-premises, including air-gapped (commercial) |
| Paid pricing | Developer: $0.002 per trace, hosted only | From $100 a month per 1.2 million spans a year, billed annually |
| Capability | Fiddler | Galileo | Prefactor |
|---|---|---|---|
| Observe | |||
| LLM call tracing | ✓ | ✓ | ✓ |
| Multi-step agent tracing (tool calls, nested spans) | ✓ | ✓ | ✓ |
| Cost tracking | Partial | ✓ | ✓ |
| OpenTelemetry support | ✓ | ✓ | ✓ |
| Evaluate | |||
| Offline evals (datasets, experiments) | ✓ | ✓ | ✓ |
| Online evals on production traffic | ✓ | ✓ | ✓ |
| LLM-as-a-judge scoring | ✓ | ✓ | ✓ |
| Code-based evaluators | ✓ | ✓ | ✓ |
| Human review and annotation | Partial | ✓ | ✓ |
| Prompt versioning and playground | — | Partial | — |
| Act on findings | |||
| Runtime controls on live agent actions | Partial | ✓ | ✓ |
| Gateway or proxy in the request path | — | — | — |
| Licence and deployment | |||
| Open source | Partial | Partial | Partial |
| Self-hosting | ✓ | ✓ | Partial |
Cisco completed its acquisition of Galileo in May 2026, and since 7 August 2026 the product has been sold as Splunk Agent Observability inside Splunk Observability Cloud. Customers onboarded before that date still use Galileo's documentation and app; later customers use Splunk's. The old Galileo pricing page redirects to Splunk, which lists annual billing and no free tier for this product.
Fiddler is independent and venture-backed, with a $30M Series C in January 2026. It sells a self-serve Developer plan alongside Enterprise contracts. If your organisation already runs Splunk, agent evaluation in the same place is a strong argument for Galileo. If it does not, Galileo is a Splunk purchase with annual terms, and Fiddler is the one you can start on without a sales call, at least for hosted observability.
Both vendors check content at run time, but they place the check differently. Galileo's Agent Control evaluates model and tool inputs and outputs during agent execution and can block prompt injection, personal data leaks and harmful content using rules set centrally, without changes to agent code. It replaced the older Protect product in June 2026, and its core was released under Apache-2.0 in March 2026, so it can be self-hosted from the open repository. Splunk documents the hosted Agent Control as an enterprise feature.
Fiddler Guardrails score requests and responses for harmful content, faithfulness, personal data and secrets in real time and hand the score to your application to act on. Blocking or redaction inline, without code changes, works when Fiddler sits behind a third-party gateway: Kong AI Gateway, AgentGateway, LiteLLM or Portkey. Fiddler's documented scope is requests and responses rather than tool-call arguments.
Neither documents approval steps for an agent's actions. If you need checks on what an agent sends to its tools, Galileo covers more of that today.
Both offer small models as lower-cost judges. Galileo has Luna-2 and, on enterprise tiers, Luna Studio for fine-tuning; Fiddler has its Centor Models and, since September 2026, a hosted Ministral 3 8B judge. One limit to check on Galileo: the Splunk-hosted edition supports five preset Luna evaluators and no custom code evaluators, which are available on-premises only. Fiddler's custom evaluators, including Python code, are available from its Developer plan.
For human review, Galileo's annotation queues became generally available in August 2026; Fiddler's span annotations are in public preview with no queue. Fiddler also monitors classic ML models for drift, explainability and fairness, which matters if the same team owns fraud or ranking models alongside agents.
Framework coverage also differs. Galileo ships Python and TypeScript SDKs and native integrations for the OpenAI Agents SDK, LangGraph, CrewAI, Google ADK, Strands Agents, Microsoft Agent Framework and multi-agent tracing over A2A, added in May 2026. Fiddler's native agent SDKs are Python only, covering LangGraph, LangChain, Strands Agents and Google ADK; other languages send OTLP. Existing Galileo users should also note that Protect was deprecated in June 2026, so any guardrails built on it need to move to Agent Control.
As of 20 October 2026, Splunk lists Agent Observability from $100 a month per 1.2 million spans a year, billed annually, including 500 million Luna tokens and five Luna fine-tuning runs a year, with no free tier listed. Fiddler's Developer plan charges $0.002 per trace and is hosted only; its free plan covers guardrails only and is reached through a demo request, and on-premises deployment is Enterprise and quote-based. Galileo counts spans and Fiddler counts traces, so model your own volumes.
Fiddler's Enterprise plan adds VPC and on-premises deployment, enterprise guardrails, a named customer success contact and custom onboarding. Splunk sells extra Luna tokens and longer data retention on top of the base price.
See Fiddler's pricing and Splunk's pricing for Agent Observability.
We sell the layer this section describes. Read it with that in mind.
Stated plainly, since Prefactor publishes this page: Prefactor is an agent evaluation platform that scores each step of a run and acts on what it finds through runtime policies on tool calls and data access, approval holds for risky actions, and a kill switch for a run, an agent or a team. Both vendors here have run-time content checks, and Galileo's Agent Control also covers tool inputs and outputs; neither documents approval holds. Prefactor is self-serve with a free tier and is not tied to Splunk, but it has no prompt playground or prompt registry and its platform is hosted, with deployment in your own environment only on enterprise terms. See the full comparison of 15 platforms.
Reviewed against public product and documentation pages on October 20, 2026. If a vendor has changed a feature, product name, or positioning since then, send a correction and we will update it. Numbered source links in the page body point to the ordered sources below.
Prefactor scores each step of every agent run and can hold or stop an action when a score says it should not happen. Start on the free tier: 5,000 spans a month, no credit card required.
Prefactor helps teams observe, evaluate, and improve their AI agents in production, across every framework and provider.