Overview

The workflow takes resumes through parsing, PII removal, hard filters and ranking, then routes remaining candidates to persona agents and human review. Recruiters can inspect ranked results and accept or reject candidates. That sequence puts deterministic screening ahead of the paid LLM stage, which can limit unnecessary model processing, but the project has no shipped accuracy benchmark to establish how well its rankings identify suitable candidates.

Resume ingestion supports PDF, DOCX and plain-text files, including multi-file uploads. Ranking uses token-cosine similarity, with technical and risk persona agents and synthesis contributing to review. Claude integration runs through Agno persona agents; an offline stub LLM can run without an API key, though it does not validate screening quality.

Key features

Hard filters and ranked review: Deterministic filters run before the paid LLM stage, and recruiters get ranked results with accept-or-reject decisions. This is a useful balance for teams wanting a reviewable workflow rather than unattended decisions.
Batch cost visibility: Users can see estimated processing costs before confirming a staged batch. This helps teams manage model spend, but does not eliminate per-token LLM costs.
Privacy and audit: CandiSift says it removes PII before candidate data reaches a model and maintains an append-only log of scores, rationales, decisions and versions. The API also has a bias-audit endpoint covering pass-through and recommendation mix. These controls support reviewability; they do not establish unbiased outcomes or ranking accuracy.
Security controls: The project describes HTTP Basic authentication, security headers, rate limiting, request-body limits, upload validation and prompt-injection defenses. Teams still need to assess deployment and security against their own requirements.
Configurable deployment: Docker deployment, SQLite by default and a configurable Postgres database URL give self-hosters a straightforward starting point and a path to use Postgres.

OCR, embeddings/FAISS and skillNER are documented swaps, not default capabilities. Organizations that depend on those capabilities need to account for implementing or integrating them rather than assuming they are included.

Pricing

Open-source self-hosted: 0.00 USD per free. There is no software fee stated, but LLM processing can incur per-token costs. This fits teams able to operate the application themselves and manage model usage. The free plan is the self-hosted software; it does not make external LLM processing free.

Platforms

CandiSift supports API, self-hosted and web use. Docker deployment and configurable SQLite or Postgres storage make it most relevant to teams that want control of their deployment and can support it operationally.

Who it's for

CandiSift is best suited to recruiting teams that want to create roles, screen uploaded resumes, review ranked candidates and record accept-or-reject decisions within a self-hosted workflow. It is a weaker fit for buyers who need validated accuracy evidence, turnkey OCR or embeddings, or a managed service rather than deployment responsibility.

Pros and cons

Pro: Hard filters precede the paid LLM stage, and staged batches show estimated processing costs before confirmation; both help teams control screening effort and model spend.
Pro: PII removal, append-only decision records and a bias-audit endpoint give teams concrete mechanisms for reviewing how screening decisions were made.
Pro: Open-source self-hosting avoids a stated software fee and offers Docker deployment with SQLite or configurable Postgres.
Con: No shipped accuracy benchmark means teams cannot use a product-provided benchmark to judge ranking quality.
Con: OCR, embeddings/FAISS and skillNER are not default capabilities, limiting out-of-box coverage for workflows that need them.
Con: Claude use can add per-token costs, and self-hosting leaves deployment operations to the adopting team.

Alternatives

For a broader candidate screening software shortlist or AI recruiting assistants options, compare tools by the workflow and deployment model your team needs.

Resume Screening AI is worth considering for per-resume pay-as-you-go screening: it charges 0.07 USD per once, includes 71 free screenings to start, and offers unlimited jobs and candidates plus up to three custom scoring fields. Its API, extension and web platforms may suit teams that do not want a self-hosted setup.
CVViZ Job Distribution may fit teams seeking a paid ATS feature set across many platforms: its Starter plan is 99.00 USD per month for five active jobs and unlimited users.
X0PA AI is another paid option with API, self-hosted and web platforms; consider it if its candidate limits and paid plans fit your requirements.
Humanly Intelligent CV Screening may suit teams wanting a web-based freemium entry point: Explorer is free and allows up to 10 CVs per screening run, with no storage.
SkipCV offers a web-based free tier with 20 credits and a maximum of eight candidates per analysis, making it an option for smaller screening batches.
Octagnt offers a free web plan for one requisition and up to 150 evaluated profiles each month; its Starter plan is 499.00 USD per month for 10 requisitions monthly.
Ducknowl uses custom pricing tailored to hiring process and screening or assessment types, with payment based on use.
Counting AI is another paid alternative.

Verdict

Choose CandiSift if your recruiting team wants an open-source, self-hosted screening workflow with cost controls and auditable human review, and has the capacity to operate it. Look elsewhere if validated ranking accuracy, default OCR or embeddings, or a managed deployment is essential.

	CandiSift	VeraHire
Free plan	Yes	No
Free trial	No	Yes
Paid from	Free	$49/mo
Open source	No	No
Platforms	api, self-hosted, Web	Web
Free plan	—	No
Paid from	—	49 /mo
Skills screening	—	Yes
ATS integrations	—	Yes
API access	—	Yes

CandiSift vs VeraHire

CandiSift

VeraHire