SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI careersnext-role

next-role

NextRole is a job search and promotion preparation tool based on GenAI and a multi-agent system. Users upload their resume and enter a target job description (JD), and the system automatically completes company research, tailored resume rewriting, interview preparation document generation, and on-the-spot cheat sheet creation. The project is driven by a multi-agent architecture and includes two types of agents: career agents and data analysis agents. It supports long-term memory, multi-turn conversational iterative editing, and runtime switching of large models. The overall tech stack (frontend, backend, Postgres, Redis, object storage, and analytics stack) can all be self-hosted in Docker, and the open-source code is provided. Users can automatically extract job information by pasting a JD link, and can progressively modify generated content through chat, making it suitable for job seekers who need to prepare efficiently for interviews and internal promotions.

Open website

About this tool

Purpose

NextRole is designed to accelerate job search and career promotion preparation with generative AI. Traditional interview preparation often requires hours of resume customization and company research. NextRole automates this process: users provide their existing resume and target job description (JD, supporting files, URLs, or pasted text), and a group of specialized AI agents takes over to complete company research, resume rewriting, round-by-round interview coaching, and generate a cheat sheet usable on the day itself. It is suitable for both external applications and internal promotion scenarios.

Main Capabilities

  • Tailored resume with PDF output: rewrites experience content based on the specific JD and company research, using rendercv to render an editable, re-renderable PDF.
  • In-depth company and role research: distills fit analysis through real-time web research.
  • Structured interview preparation: generates a self-introduction and round-by-round STAR stories mapped to the role.
  • Same-day cheat sheet: generates a one-page PDF cheat sheet per round for last-minute high-pressure review.
  • Time-boxed preparation plans: can adapt to study plans of 1 month, 2 weeks, or even just 3 hours.
  • Paste JD URL: point to a recruiting page to automatically extract and process that job information.
  • Conversational iteration: make modifications through chat such as "add a 4th round" or "add React to skills", with streaming multi-turn editing executed by the agent responsible for the corresponding file.
  • Built-in workspace: supports upload, preview (PDF / MD / YAML / JSON / code), print to PDF, and runtime LLM switching.
  • Data analysis agent: can answer questions about the product itself (activity, reliability, LLM cost, agent behavior), querying a read-only repository and presenting answers with charts that persist across the conversation.

How to Use

The official hosted version next-role.dev is coming soon; until then, you can self-host the deployment. The entire tech stack (frontend, backend, Postgres, Redis, S3-compatible object storage, and an analytics stack composed of ClickHouse, Dagster, Cube, and Superset) runs through Docker. Users can set the models used by the main agent and sub-agents in the in-app Configuration dialog, using the provider:model string format (for example, anthropic:claude-sonnet-5), without rebuilding; leaving it blank uses the default values, and settings are saved in browser local storage.

Multi-agent architecture based on a self-hosted agent server built with LangGraph + DeepAgents. The career agent handles onboarding, document processing, and the final cheat sheet, and delegates research, resume customization, and interview coaching to three declarative sub-agents (defined in subagents.yaml, each with its own model, tools, and skills). The backend executes agents; core-server manages metadata and a persistent run queue in PostgreSQL; checkpoints and memory use PostgreSQL directly; Redis handles queue signals and real-time events; uploaded files, PDFs, and diagrams are persisted in S3-compatible object storage. The career preparation workflow is divided into five stages: onboarding, document processing, research, customization and coaching (executed in parallel), cheat sheet generation, followed by support for continuous multi-turn editing.

On the security side, risky execute (bash) commands pause in the UI and wait for approval/edit/rejection, enabled by default; SANDBOX_PROVIDER can be used to choose whether commands execute on the local backend host or in an isolated remote micro-VM (E2B / self-hosted CubeSandbox). The default is zero-login single-user mode, and multi-user mode can also be enabled (supporting Google OAuth and/or email password, based on self-hosted Better Auth), providing user isolation for threads, files, and memory. Secrets are stored only in .env (already gitignored), and gitleaks runs on every commit.

Use cases

Suitable for candidates who are job hunting and need to quickly customize resumes and prepare for interviews for different JDs; also suitable for employed individuals seeking internal promotion who need structured preparation materials. For teams or individuals who care about data privacy and want to self-deploy and control their data, the self-hosted solution provides the required infrastructure.

Limitations

The official hosted version is not yet available, so self-hosting is required; the full analytics stack requires an additional roughly 2-4 GB of memory. Output quality depends on the chosen model, and smaller local models trade quality for zero cost. When using the data analytics agent to read all user activity, it must be explicitly allowed via ANALYTICS_AGENT_ALLOWED_USERS (when authentication is enabled; an empty value denies everyone, and single-user deployments are unaffected). Product analytics (PostHog) and LLM tracing storage (Langfuse) are considered follow-up work in the plan.

Related tools

Browse category