Keep your Weave calls. Add security and compliance.
Weights & Biases Weave is a great LLM tracing + evaluation platform. EvalGuard imports your Weave calls — op names, token usage (OpenAI and Anthropic naming), cost, and exceptions all carry over — then adds hosted red-team scans (300+ plugins), a runtime LLM firewall, an AI gateway, and a SOC 2 evidence engine on the same data. No sign-up needed to run your first import.
Honest positioning
Where Weave stops, EvalGuard keeps going
Weave is a first-class tracing + eval tool with a clean call tree and the W&B ecosystem behind it. EvalGuard overlaps on tracing and evals, then extends into the hosted security platform, runtime firewall, AI gateway, and compliance work you'd otherwise buy separately.
| Capability | W&B Weave | EvalGuard |
|---|---|---|
| LLM call tracing (trace tree) | Yes | Yes — OTLP + OpenInference ingestion, governed |
| Evaluation + scorers | Yes | Yes — 200+ scorers (LLM-as-judge, pairwise, rubric) |
| Native token / cost tracking | Yes | Yes — per-org cost ledger + budgets |
| Open source & self-hosting | OSS SDK, hosted backend | Yes — Apache-2.0 core, self-host available |
| Red-team & security scans | Not offered | Yes — 300+ attack plugins with threat-feed sync |
| Runtime LLM firewall / guardrails | Not offered | Yes — real-time input + output firewall |
| AI gateway / BYOK proxy | Not offered | Yes — BYOK gateway, 15 proxied providers, semantic cache |
| SOC 2 evidence engine | Not offered | Yes — live evidence engine + audit log |
Migration path
Bring your calls in one command
Everything stays local until you choose to run it — we never touch your W&B project. Export your calls, then convert them to neutral-shape spans with the EvalGuard CLI.
# Python: calls = client.get_calls(include_costs=True)
# pd.DataFrame([c.to_dict() for c in calls]).to_json('weave-calls.json', orient='records')npx @evalguard/cli import:traces --from weave weave-calls.json --output spans.jsonnpx @evalguard/cli scan # 300+ red-team attacks on your migrated dataToken usage keyed by model, per-model cost (with include_costs), and exceptions all map over — a failed call becomes an error span, a normal one stays ok. Then layer on red team, a runtime firewall, and a SOC 2 evidence engine.
Want a hand with the migration?
Send us your Weave export and we'll help you map it and validate the first import. Free.