Openings

A self-hosted job search workspace. Collect offers, keep your career knowledge and templates together, and let your own AI agent prepare applications that you review and submit. Jobs, documents and file bytes live in one portable SQLite database.

Start with Docker · Connect an agent · Documentation · Changelog

Features

• Collect: scheduled keyword/location searches on LinkedIn, Indeed and Glassdoor, or selected company boards (Greenhouse, Lever, Ashby, Workday and more) and RSS feeds. Role, location and exclude_titles filters skip noise before it reaches your Inbox. You can also add any job by URL or pasted text.

• Triage: review the Inbox, keep jobs on a Shortlist or dismiss them; dismissed jobs stay remembered. Possible duplicates groups one vacancy collected from several sources so you can merge verified copies; bulk actions work on up to 200 jobs.

• Decide once: a Policy of short statements holds your rules for shortlisting and dismissing, the facts employers ask about and your writing rules. Your agent proposes items; they count after you confirm them.

• Prepare: store your Profile, Career entries, approved claims with typed selection fields, LaTeX templates with typed rules and writing instructions. Each application freezes the exact context its CV, cover letter and form answers were written from, with PDF/source revisions and review.

• Plan: give applications a priority, a deadline and the employer's requirements (authorship rules, attachments, references, language, eligibility). Filter by readiness (blocked, no documents, missing PDF, needs review, reviewed, ready) and sort by deadline or priority. Close, reopen or delete applications in bulk.

• Hand over: an application is marked Ready once its documents are approved and nothing blocks it. Ready to apply is your queue, and a later change to the documents clears the mark.

• Stay consistent: applications built on older inputs are flagged after Profile, template or Career changes, and review checks a document's content against its .tex source.

• Track: move applications through Preparing, Applied, Interview, Offer and Closed with recorded outcomes, history and a progress funnel.

• See what the agent did: every change records who made it. An agent works in runs with a stored report, and asks you questions that you answer under Activity.

• Tell the agent what to do next: write a request under Activity, or Ask your agent on a job or an application. The agent picks it up at its next run and reports the result.

• Automate: the dashboard, REST API and MCP share one service layer and the same operations.

Related MCP server: Jobs MCP Server

Agent-operated

Openings works best when your own AI agent runs it. The server ships workflow skills and serves them to any connected agent (skills_list, skill_get and one MCP prompt per skill). With them, an agent reviews new jobs against your policy, prepares a CV and letter for every shortlisted job, records employer requirements, tunes searches and keeps your Career record current. You answer its questions, apply to the prepared Drafts and mark them Applied.

The agent uses its own model access and tools and never submits applications for you. Read Agent-operated job search.

Start with Docker

Requires Docker with Compose v2. Published images support Linux amd64 and arm64.

git clone https://github.com/VincenzoImp/openings.git
cd openings
cp .env.example .env
docker compose pull
docker compose up -d --no-build

Open http://127.0.0.1:8080, choose an owner password and add your first job in Jobs. Set up Collection when you want scheduled searches; new sources start paused. Build your Career and Templates library before preparing documents.

The Compose limit is 0.5 CPU / 512 MiB; allow host resources beyond this container budget. Setup and first use covers LAN access, storage and upgrades.

Connect an agent

1. In Settings → Agent access, create an agent token and copy the configuration for Claude Code, Codex or another MCP client. An agent token cannot change your password, manage tokens or restore a backup. The endpoint is http://127.0.0.1:8080/mcp with Authorization: Bearer <token>.

2. Ask the agent to load the setup-openings skill to onboard your career, templates and first searches.

3. For a persistent setup, copy the operator workspace template: it holds the operating mandate and generates MCP configuration for six clients.

Agents and MCP explains the tools, transfers and state rules.

Documentation

Collection · Career · Applications · Agents and MCP · REST API · Operations

Project

Contributing · Code of conduct · Architecture · Security · Changelog · MIT license