0.121.14 (2026-06-02)
Features
- add A2A provider (#9586) (963b264)
- assertions: add agent-rubric grader (#9453) (cadb3c5)
- build: publish a lightweight promptfoo/contracts subpath (#9535) (6e89fd4)
- cli: add seeded random sampling (#9522) (6e4ba60)
- eval: extract eval-creator readiness and validation modules (#9397) (d3af118)
- eval: surface trace linkage on result rows (#9027) (9468d66)
- providers: add OpenAI GPT-5.5/5.4 frontier models + Codex on Amazon Bedrock (#9587) (1e35267)
- providers: add OpenAI originator header (#9474) (6141070)
- providers: expose traceable agent turn markers (#9475) (4018837)
- providers: promote Fireworks AI from registry stub to dedicated provider (#9542) (2bae148)
- tracing: add runtime receiver controls (#9028) (90df6a1)
Bug Fixes
- assertions: don't classify gen_ai.tool.definitions chat spans as tool calls (#9524) (b59f397)
- cli: preflight force imports before collision lookup (#9570) (0b93733)
- cloud: on-prem API host for guardrails and http-generator, with host-resolution tests (#9580) (b9a014a)
- cloud: use on-prem API host in checkEmailStatus (#9576) (063c62b)
- db: avoid SQLITE_LOCKED flakiness in shared-cache test database (#9567) (1fdb59b)
- db: serialize libsql test database cleanup (#9540) (f4380c2)
- deps: keep ModelAudit pydantic-core pinned to compatible 2.46.4 (b2b35b0)
- deps: update dependency ai to ^6.0.190 (#9577) (9ec614a)
- eval: canonicalize retry JSONL output with atomic rewrites (#9547) (8d7c920)
- eval: redact credentials from the persisted browser store (#9396) (4d5bed5)
- evaluator: preserve and harden programmatic JSONL output (#9538) (8ddd906)
- output: redact api-key and legacy transport headers in JSONL/DB (#9546) (e194c85)
- providers: handle Codex SDK rate limits (#9473) (76d3db4)
- providers: inject n8n sessions into custom bodies (#9527) (9cc0542)
- providers: preserve n8n array body templates (#9544) (6cdf63d)
- providers: preserve streamed Anthropic refusal guardrails (#9560) (ff8eafd)
- providers: serialize persistent browser sessions (#9414) (097ff9b)
- redteam: authenticate remote-generation requests against on-prem cloud (#9584) (7df8fae)
- util: restore nested SAS tokens after array reorder (#9528) (9759e5a)
- webui: render negative-only metric charts (#9526) (7ccafa4)