Hyper-personalized job search that filters postings based on what matters to you: role scope, autonomy, management experience, fit with your background, and more.
- Important details are often missing from filter fields or buried in job descriptions.
- Reposts and duplicates flood search results with irrelevant and repetitive postings.
- Traditional search requires you to compress a nuanced decision into a few keywords. Results often miss what matters: the work itself, decision-making authority, management experience, and fit with your background.
- After applying filters and getting search results, the real work begins: reading posting after posting to find the few roles you would seriously consider.
- Search by fit, not just keywords: describe your preferences in plain English, including concrete constraints and less searchable qualities such as autonomy, role scope, and access to company leadership.
- Let your agent do the filtering: your agent reviews every new posting against your preferences, filters out clear mismatches, and reads promising job descriptions in full. It evaluates the work behind the title, surfacing relevant roles at unfamiliar companies or under titles you might not have thought to search for.
- Know why each job fits: get a concise explanation of why each role is a strong, moderate, or weak match, so you can quickly focus on the opportunities that best fit your preferences, background, and interests.
- Catch new matches early: schedule your search to run daily, hourly, or on your preferred cadence, so new postings are automatically evaluated as they appear.
-
Install Job Search for your coding agent.
-
Start a conversation with:
Set up my job search. I'm looking for a senior product design role, remote in the US, with end-to-end ownership. I don’t want brand-heavy or solo-designer roles. My background is primarily B2B SaaS...
The agent then:
- Checks its prerequisites
- Documents your preferences, and
- Searches live postings.
If useful roles are available, it shows only the ones matching your preferences.
A sentence or two is enough to begin. You can also share relevant material, such as a resume, cover letter, or notes from previous applications to use as it evaluates job opportunities.
You: Run a search now.
Job Search: Searching for new roles matching your preferences...
Found 42 postings. 9 are new. Reading the promising ones in full...
Here are the first strong matches while I keep reviewing the rest:
• Senior Product Designer — Tidewater Health — Remote (US)
Owns a care-navigation area end to end; the healthcare mission you're after.
...still reviewing the remaining postings...
Job search digest — 2026-06-05
9 new postings · 2 strong · 2 moderate · 2 weak · 3 filtered out
See the complete sample digest.
| Goal | Example |
|---|---|
| Start | "Set up my job search. I want a senior product design role, remote in the US, with end-to-end ownership. I don’t want brand-heavy or solo-designer roles. My background is primarily B2B SaaS...” |
| Refine preferences | “Make working closely with product and engineering a strong preference” |
| Search now | “Run a search now,” or “check for new jobs.” |
| Schedule searches | “Run my search every morning.” |
| Review a match | “Why is this a strong match?” or “Does this posting fit what I want?” |
| Pause or stop | “Pause my schedule,” or “stop scheduling.” |
Job Search gets live postings through the agent-data command-line tool. Onboarding handles the setup:
- If the CLI is missing, the agent offers to install it with
npm install -g agent-data. - If authentication is missing, the agent helps you create an API key and verify authentication.
Agent-data offers a 100-call monthly free tier (no credit card required). Typical runs consume ~10-30 calls. See what a run spends or your billing page for additional details.
Currently supported sources: LinkedIn Jobs, Ashby, Greenhouse, and Lever.
Choose your coding agent below. After installation, use the Quickstart sentence. Installation succeeded when Job Search opens onboarding and begins the agent-data check.
Run these commands inside Claude Code:
/plugin marketplace add agent-data/job-search
/plugin install job-search@agent-data
Use /job-search if you need the command shortcut.
Run in your shell, then launch or restart Codex:
codex plugin marketplace add agent-data/job-search
codex plugin add job-search@agent-data
codexUse $job-search if you need the skill shortcut.
Clone the repository, then symlink it into Cursor's local plugins directory:
git clone https://github.com/agent-data/job-search
ln -s "$(pwd)/job-search" ~/.cursor/plugins/local/job-searchAdd Job Search to the plugin array in your global OpenCode config at
~/.config/opencode/opencode.json, or to an opencode.json in a project where you want to use it:
{
"$schema": "https://opencode.ai/config.json",
"plugin": ["job-search@git+https://github.com/agent-data/job-search.git"]
}Restart OpenCode, then start with: “Set up my job search.” OpenCode installs the git package with Bun and the plugin registers all seven bundled skills. If your config already has plugins, add this entry to the existing array rather than replacing it.
See .opencode/INSTALL.md for verification, revision pinning, local development,
and troubleshooting.
gemini extensions install https://github.com/agent-data/job-searchcopilot plugin marketplace add agent-data/job-search
copilot plugin install job-search@agent-datadroid plugin marketplace add https://github.com/agent-data/job-search
droid plugin install job-search@agent-datapi install git:github.com/agent-data/job-searchFor an editable local install, run pi -e /path/to/job-search.
Copy and paste the following into a new Hermes Agent session:
Retrieve and follow the installation instructions at:
https://raw.githubusercontent.com/agent-data/job-search/main/INSTALL_FOR_HERMES.md
After installation, start a new Hermes Agent session and use the Quickstart sentence above.
Compatibility details
These expectations apply to every listed agent:
| Field | Expected support |
|---|---|
| OS / architecture | macOS and Linux; arm64 and x86_64 |
| Recurring scheduler | cron or launchd, with a session loop fallback |
| Modes | interactive and background (headless) |
Job Search is compatible with Claude Code, Codex, Cursor, opencode, Gemini CLI, GitHub Copilot CLI, Factory Droid, Pi, and Hermes Agent.
Every run — the one you ask for and the one a schedule starts — reads and judges postings on the model of the session it runs in. Nothing pins a separate model for the search. A recurring schedule is recorded only after the agent runs the real scheduled command once and confirms that it reached agent-data and wrote the workspace.
If something fails, ask “why did my last run fail?” or “how does my job search agent work?” The agent reads the run's record and the digest it wrote, both already on your machine, and tells you what stopped the run and what fixes it. To report a bug, copy what it tells you into a GitHub issue.
Start with AGENTS.md. It points to the architecture, contributor workflow, test guide, and the canonical runtime contracts in job-search-runbook and agent-data-reference.