AI-assisted HR onboarding system for Feishu/Lark featuring document OCR, YAML workflows, and human review gates.
Automate document extraction and Bitable sync while keeping HR staff in control of key decisions.
HR Onboarding Agent coordinates employee onboarding within Feishu and Lark using FastAPI and SQLite. It provides separate workspaces for candidates and HR teams, relying on deterministic rules and manual approval gates to review uploaded candidate documents.
The tool ingests candidate documents, runs OCR and optional LLM passes to extract information, and updates Feishu/Lark Bitable records. You configure workflow steps, due dates, and owner roles using safe YAML files, while APScheduler manages automated reminders for overdue tasks.
It suits HR and People Operations teams using Feishu or Lark who want automated document parsing without giving up manual control over candidate approvals. It is not suitable for organizations needing a full legal payroll engine, out-of-the-box SSO, or multi-tenant hosting out of the box.
Define onboarding stages, due dates, owner roles, and time offsets in strict versioned YAML files.
Inspect uploaded files against file size limits, magic bytes, and MIME types before executing OCR processing.
Flag suspicious self-approval text or prompt injection attempts, forcing mandatory human HR verification.
Update candidate status cards and synchronize employee metadata directly into Feishu or Lark Bitables.
Connect optional external models like OpenAI, Gemini, Doubao, Zhipu, or DeepSeek for second-pass document reviews.
HR Onboarding Agent is free and open source under the Apache-2.0 license. There are no paid hosted tiers provided by the authors, so your expenses consist entirely of self-hosting infrastructure and optional LLM/OCR API usage.
880 open remote AI-training roles, updated November 15, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
These reviewed AI tools do the same job instantly — from our scored directory.
Yes, you can run the application entirely with the built-in mock provider without making external API calls.
It treats all uploaded files and OCR output as untrusted data, applying deterministic security checks that trigger mandatory human HR review when suspicious text is detected.
No, current versions require separate deployments for each tenant and lack multi-tenant data isolation.
See Zekai first in Google