Skip to content
DeepEval → EvalGuard

Keep your metrics. Add security and guardrails. 

DeepEval is a great open-source, pytest-style metrics library — GEval, AnswerRelevancy, Faithfulness, Hallucination, and the rest. EvalGuard imports both your evaluate() results and your golden datasets, then adds hosted red-team scans (300+ plugins), a runtime LLM firewall, and a SOC 2 evidence engine on the same data. No sign-up needed to run your first eval.

SOC 2 evidence engineISO 42001 mappedEU AI ActGDPR

Honest positioning

Where DeepEval stops, EvalGuard keeps going

DeepEval is a first-class open-source metrics library — the cleanest pytest-style eval developer experience out there, and the metrics are genuinely good. EvalGuard overlaps on metrics, goldens, and CI, then extends into the hosted security, runtime firewall, and compliance work you'd otherwise buy separately.

CapabilityDeepEvalEvalGuard
Open-source metrics (GEval, AnswerRelevancy, Faithfulness, Hallucination) YesYes — 200+ scorers (LLM-as-judge, pairwise, rubric)
Pytest-native / CI integration YesYes — `gate` command + CI action, fail-closed
Golden datasets & synthetic data YesYes — versioned datasets + test generators
Open source & self-hosting YesYes — Apache-2.0 core, self-host available
Red-team & security scansVia DeepTeamYes — 300+ attack plugins
Hosted dashboards & RBACConfident AIYes — governed dashboards, roles, audit log
Runtime LLM firewall / guardrailsNot in the OSS package — a Confident AI featureYes — real-time input + output firewall
AI gateway / BYOK proxyConnects to OpenRouter / Portkey / LiteLLM; doesn't ship oneYes — BYOK gateway, 15 proxied providers, similarity response cache
SOC 2 evidence automationNot a documented product featureYes — live evidence engine + audit log

DeepEval claims on this page were verified against their source github.com/confident-ai/deepeval @ bd10fa61c (v4.1.5, 2026-08-05) — on 2026-08-10. The “Not offered” row was settled by searching that tree for guardrail modules and classes by filename and by content — there are none, and their own README still lists Guardrails as an unchecked roadmap item. Capabilities move; check their current docs before you decide. “SOC 2 evidence automation” describes a product capability, not any vendor’s own certification status.

The red-team cell is their own wording: the deepeval/red_teaming/package contains one file, a README reading “The Red Teaming module is now in DeepTeam for deepeval-v3.0 onwards”. Hosted dashboards — and guardrails — are Confident AI, their commercial platform, not the Apache-2.0 package.

Migration paths

DeepEval is two things. So the move is two paths.

Everything stays local until you choose to run it — we never touch your DeepEval account or Confident AI project. Convert what you export with the EvalGuard CLI.

A

Results — bring your evaluate() runs

Serialize a DeepEval EvaluationResult (its test_resultslist) to JSON, then convert it to neutral-shape spans — every metric's score, threshold, pass/fail, and reason maps over. A failed assertion stays a normal run; only a real evaluation error is flagged as an error.

# evaluate() results → neutral spans (metrics fanned out, no timing invented):
npx @evalguard/cli import:traces --from deepeval deepeval-results.json --output spans.json
B

Golden datasets — convert and run

Export a DeepEval golden dataset, convert it to a runnable EvalGuard config — inputs, expected outputs, and RAG context all carry over — then run it keyless on your machine, or on the EvalGuard cloud.

1
Export your dataset from DeepEval
# Python: dataset.save_as('json', directory) → a JSON list of goldens
2
Convert it to an EvalGuard config
npx @evalguard/cli import:deepeval deepeval-dataset.json -o evalguard.config.json
3
Run it keyless — no account, no API key (echo provider)
npx @evalguard/cli eval:local evalguard.config.json --provider echo

Your DeepEval metrics (GEval and friends) are Python code, so they don't travel inside a dataset export — the import attaches a runnable equals scorer when your goldens carry an expected output, and points you at llm-grader / semantic-similarity to re-create a GEval-style rubric. The echo provider runs the whole loop with no API key — swap in --provider openai (with your key) for real model calls, or run on the cloud for shared dashboards and run history:

# Run on the EvalGuard cloud (log in first):
evalguard eval --project <id>

Want a hand with the migration?

If you have a large golden dataset or a suite of custom GEval metrics, send us your DeepEval export and we'll help you map it and validate the first run. Free.