Fireworks AI

OutdatedAudit Done
View Session

Health

Outdated

Benchmark

14/100

SDK Version

>=1.7.0

Latest: 2.12.1

Audit Status

Audit Done

Started May 11, 2026

Approval

Not requested

Latest Audit Result

Audited exa-labs/exa-fireworks at f747c36. The integration is a Fastest RAG reference stack using Exa only through exa.search_and_contents in fastest-rag/src/fastest_rag/pipeline.py, with text extraction enabled and highlights disabled; Fireworks handles answer generation and streaming. Declared search capability is implemented, but declared metadata is stale/incomplete: the accessible repo and key files differ from the prompt, search types actually exposed are neural/keyword/auto with neural default, and the exa-py dependency is >=1.7.0 while PyPI latest is 2.12.1. Changelog review found /research, startCrawlDate/endCrawlDate, resolvedSearchType, and highlightScores sunset/removal items; this repo does not use those deprecated/sunset features.

Health: outdated
Current SDK: >=1.7.0
Latest SDK: 2.12.1

Missing Features (2)

  • Declared repo/key files mismatch remains: accessible repo is exa-labs/exa-fireworks with fastest-rag/pyproject.toml; no top-level main.py or requirements.txt exists.
  • Declared search type coverage is incomplete/stale: implementation exposes neural, keyword, and auto with neural default, not only auto.

Marked outdated since May 11, 2026

Missing Features (8)

  • Declared repo/key files mismatch remains: accessible repo is exa-labs/exa-fireworks with fastest-rag/pyproject.toml; no top-level main.py or requirements.txt exists.
  • Declared search type coverage is incomplete/stale: implementation exposes neural, keyword, and auto with neural default, not only auto.
  • Update the exa-py dependency constraint from >=1.7.0 to a current 2.x range; PyPI latest verified as 2.12.1.
  • Update declared repo/key-file metadata: current accessible repo is exa-labs/exa-fireworks with fastest-rag/pyproject.toml and fastest-rag/src/fastest_rag/*.py; no top-level main.py or requirements.txt exists.
  • Update declared search type coverage: code exposes neural, keyword, and auto via RAGConfig.search_type and defaults to neural, while declared capabilities list only auto.
  • Consider switching the default search_type from neural to auto to match current Exa default guidance, unless preserving neural scores is intentional.
  • Expose content extraction controls for the existing search flow instead of hardcoding text=True and highlights=False; current Exa guidance recommends choosing text/highlights/summary per use case and supports maxCharacters/maxAgeHours controls.
  • Stop reading result.score as a stable field for auto searches if the integration switches/defaults to auto; Exa changelog says score is removed for auto search but remains for neural search.