CandiSift
github.com
- Free plan
- Web
- API
- Self-hosted
What CandiSift is
CandiSift is a free, open-source applicant tracking system for screening resumes, parsing CVs, matching candidates to job descriptions, and ranking results with a bias-audit workflow. Recruiters can create roles, upload multiple resumes in PDF, DOCX, or plain text, then review ranked candidates and accept or reject them. Its process removes personally identifiable information before model processing, applies deterministic hard filters, ranks candidates, and sends those remaining to persona agents and human review. Token-cosine ranking, technical and risk persona agents, synthesis, and the recruiter interface support that review. Before confirming a staged batch, users can see estimated processing costs; hard filters run before the paid LLM stage. CandiSift connects to Anthropic Claude through Agno persona agents, or uses an offline stub LLM without an API key. It can run in Docker, stores state in SQLite by default, and supports a Postgres database URL. The project describes authentication and other security measures, along with an append-only audit log and an API endpoint for bias audits. It has no shipped accuracy benchmark; OCR, embeddings/FAISS, and skillNER are documented swaps rather than included default capabilities. LLM processing can incur per-token costs.
Who it suits
CandiSift suits recruiters who want a self-hosted workflow to screen uploaded resumes, review ranked candidates, and make accept-or-reject decisions.
Good to know, and what to watch
- Accepts PDF, DOCX, and plain-text resumes.
- Removes personally identifiable information before model processing.
- Shows estimated processing costs before batch confirmation.
- Supports SQLite by default and a Postgres database URL.
- The project has no shipped accuracy benchmark.
- LLM processing can incur per-token costs.
- OCR, embeddings/FAISS, and skillNER are not default capabilities.
Plans
Every plan and priceOpen-source self-hosted
Open source
The verdict
CandiSift combines resume parsing, filtering, ranking, and recruiter review in an open-source workflow. Consider its missing accuracy benchmark and possible LLM costs when assessing whether it fits a hiring process.
What the maker publishes
- Bias audit
- The API includes a bias-audit endpoint covering pass-through and recommendation mix. source
- Cost controls
- Users can see an estimated processing cost before confirming a staged batch, and deterministic hard filters run before the paid LLM stage. source
- Deployment and storage
- The app can run in Docker, stores state in SQLite by default, and supports configuring a Postgres database URL. source
- File formats
- Resume ingestion supports PDF, DOCX, and plain-text files, including multi-file uploads. source
- Intended users
- The README describes a recruiter workflow for creating roles, screening uploaded resumes, reviewing ranked results, and accepting or rejecting candidates. source
- LLM integration
- CandiSift integrates with Anthropic Claude through Agno persona agents; without an API key, it can use an offline stub LLM. source
- Notable limits
- The project says it has no shipped accuracy benchmark and that OCR, embeddings/FAISS, and skillNER are documented swaps rather than included default capabilities. source
- Privacy and audit
- CandiSift says it strips PII before a model sees candidate data and keeps an append-only audit log of scores, rationales, decisions, and versions. source
- Product
- CandiSift is an open-source applicant tracking system for AI resume screening, CV parsing, job description and candidate matching, and bias-audited ranking. source
- Ranking and review
- The product provides token-cosine ranking, technical and risk persona agents, synthesis, and a recruiter interface with ranked results and accept or reject decisions. source
- Screening flow
- Its workflow parses resumes, strips personally identifiable information, applies hard filters, ranks candidates, and sends survivors to persona agents and human review. source
- Security
- The repository describes HTTP Basic authentication, security headers, rate limiting, request body limits, upload validation, and prompt-injection defenses. source
Checked against the maker's pages on 4 Oct 2026.