GONATURELLE vs Arize Phoenix: open-source tracing and evals or a regulated eval gate?
An engineering comparison for teams shipping AI agents into regulated workflows. Reviewed November 2026.
Arize PhoenixCapability rows judged against public documentation, November 2026. No vendor prices quoted.
The short version
Arize Phoenix is a source-available AI observability and evaluation tool, released under the Elastic License 2.0, that you can run in a notebook, in Docker or on Kubernetes without a license fee. It is built on OpenTelemetry and the OpenInference conventions, with auto-instrumentation for most major agent frameworks and model SDKs. Phoenix gives you trace inspection, an evals library with LLM and code evaluators and prebuilt templates (hallucination, relevance, QA correctness and more), versioned datasets, experiments and a prompt playground. Arize also sells a commercial platform, Arize AX, for teams that want a managed service. For a team that wants open, self-hosted tracing and flexible evals, Phoenix is an excellent starting point. GONATURELLE does not compete on tracing. It adds what Phoenix leaves to you in regulated work: expert-labelled vertical datasets with published agreement, calibrated judge ensembles, trajectory assertion primitives, and a CI gate that produces signed audit evidence.
Where Arize Phoenix is the better choice
- License and cost to startSource-available under Elastic License 2.0; free to self-host.
- InstrumentationOpenTelemetry native with OpenInference auto-instrumentation for many frameworks and providers.
- Trace inspectionFull trace and span inspection for development and production traffic.
Capability by capability
Filter by who is stronger. Rows we mark in the alternative's favour are the ones we would want to know about as buyers.
| Capability | GONATURELLE | Arize Phoenix | Edge |
|---|---|---|---|
| License and cost to start | Commercial; 14-day trial with 5,000 runs, Developer plan at $499 per month. | Source-available under Elastic License 2.0; free to self-host. | Arize |
| Instrumentation | Python and TypeScript SDK decorator; accepts OpenTelemetry traces as BYOD input. | OpenTelemetry native with OpenInference auto-instrumentation for many frameworks and providers. | Arize |
| Trace inspection | Trajectory diff timeline for evaluation runs only. | Full trace and span inspection for development and production traffic. | Arize |
| Eval templates | Domain rubrics for clinical, credit, legal, claims and pharmacovigilance tasks. | General LLM eval templates (hallucination, relevance, toxicity, QA correctness) plus custom evaluators. | Parity |
| Expert-labelled datasets | Vertical packs double-labelled by credentialed reviewers with adjudication and per-pack agreement. | Datasets you curate from traces or upload; labels come from your team. | GONATURELLE |
| Judge agreement evidence | Kappa and AC1 per judge against expert consensus, with bias controls (order swap, length normalization, cross-family judges). | You can measure evaluator agreement against your own labels; no expert-calibrated judges ship. | GONATURELLE |
| Trajectory assertions | Declarative order, schema, bound, hallucinated-argument and loop assertions. | Span data is available for custom evaluators; no dedicated assertion engine. | GONATURELLE |
| CI gate | Hosted gate action with fast path, full suite, PR comments and required status checks. | Experiments can be scripted into CI with the Python client; gating logic is yours to build. | GONATURELLE |
| Audit evidence | SHA-256 signed manifests and regulatory summaries per gated run. | Not a focus of the project. | GONATURELLE |
When to choose each
Choose Arize Phoenix when
- You want open, self-hosted tracing with no vendor contract and are comfortable operating it yourself.
- OpenTelemetry is your standard and you want auto-instrumentation across frameworks with no code changes.
- You are in research or early development and need flexible evals over traces more than a release gate.
- Budget is the binding constraint and your domain has no regulatory review step.
Choose GONATURELLE when
- You need ground truth from physicians, attorneys or actuaries and cannot produce enough of it internally.
- Your release process requires a gate with thresholds, required checks and an evidence trail, not a notebook.
- You need to show reviewers why your judge is trustworthy, with agreement statistics per rubric.
- You run on-prem or air-gapped and want a supported, signed evaluation pipeline with a BAA.
Running both
Most teams that evaluate GONATURELLE already run Arize Phoenix. This is the setup they settle on.
- 01
Instrument once with OpenInference and send spans to Phoenix for development and production tracing.
- 02
Point the GONATURELLE OTLP importer at the same exporter, or export spans from Phoenix, to turn selected traces into regression cases.
- 03
Use Phoenix experiments for open-ended exploration and GONATURELLE suites for the cases that decide whether a pull request merges.
- 04
In air-gapped installs both run inside the same Kubernetes cluster, so no trace data crosses your network boundary.
Migration steps
Most teams finish in one sprint, with both systems running in parallel until results line up.
Read the eval guide- 1Keep your OpenInference instrumentation; add an OTLP exporter to the GONATURELLE collector endpoint, or batch export spans as JSONL.
- 2Convert Phoenix datasets to an GONATURELLE BYOD suite with gonaturelle datasets import --format openinference.
- 3Re-run your Phoenix eval templates as plugin assertions and compare them with GONATURELLE calibrated judges on the same cases.
- 4Wire gonaturelle/gate-action@v2 into the pipeline and make it a required status check.
Frequently asked questions
Is Arize Phoenix open source?
Phoenix's source is public and it is free to self-host. It is licensed under the Elastic License 2.0, which is source-available rather than an OSI-approved open-source license, so check the terms if you plan to offer it as a managed service.
Can GONATURELLE read OpenInference traces?
Yes. GONATURELLE accepts OpenTelemetry spans, including OpenInference semantic conventions for LLM, tool and retriever spans, and maps tool spans to trajectory steps so assertions run against them.
Why not build the gate on Phoenix ourselves?
Many teams do, and it works when the cases and judges are good. The cost is in the parts that are not software: recruiting credentialed reviewers, adjudicating disagreements, measuring judge agreement and keeping packs current as regulations change. GONATURELLE sells that work as versioned packs.
What about Arize AX?
Arize AX is Arize's commercial platform built around observability and evaluation at production scale. The same split applies: AX watches deployed behaviour, GONATURELLE gates releases on regulated test suites.
Other comparisons
GONATURELLE vs LangSmithProduction observability vs CI regression gate
GONATURELLE vs BraintrustExperiment workflow vs regulated golden sets
- GONATURELLE vs Manual prompt testingSampled vibe checks vs every-PR coverage
- GONATURELLE vs Generic open-source eval toolsGeneric heuristics vs domain golden datasets
Keep Arize Phoenix. Add the gate.
Run your agent against an expert-labelled regulated pack in minutes. 14-day free production trial with 5,000 runs, no card.