Your coding agent, turned into a job-search partner. Every morning it finds startups that just raised money, checks whether they fit you, researches the company and the founders, and drafts a short, specific email for you to send. It never sends anything itself.
Works with any CLI coding agent: Claude Code, OpenCode, Codex, Gemini CLI, Pi, or anything that reads AGENTS.md.
I built this for my own job search. People asked for it on X, so here it is. @omkarchebale
| Workflow | What you get |
|---|---|
| Funding sweep (daily) | Startups funded in the last 24 hours that match your roles, locations and visa limits. Each with a sourced funding line, live open roles, the right person to contact, an email that's verified or clearly marked as a guess, a fit score, and a draft email |
| X / LinkedIn hiring sweep (daily) | Founder "we're hiring" posts from the last few days, the role checked on the company's own careers page, and a DM written to match exactly what the post asks for |
| Research a company | A full brief: what they build, funding, founders' track records, open roles, open-source entry points, red flags, and a verdict |
| Follow-ups | Checks your inbox for replies, then drafts the follow-ups that are due |
Everything lands in plain markdown files: leads.md as your tracker, companies/ for briefs, outreach/queue/ for drafts.
A company that raised money last week is about to hire, often before it posts a job, and before hundreds of applications arrive. Job boards are the fallback here, not the plan. See playbook/strategy.md.
git clone https://github.com/Chebaleomkar/job-radar
cd job-radarOpen your coding agent in the folder and say:
Read SETUP.md and set me up.
It checks what you have installed, connects the tools, interviews you to build your profile, and does a test run. Or skip the clone: paste this repo's link into your agent and say the same.
| Thing | Needed? | Cost |
|---|---|---|
| A coding agent + a model | Yes | Bring your own. Use your existing plan or credits, or free models (Gemini CLI login, OpenRouter :free models, OpenCode free models). We don't provide credits |
| Exa + Parallel Search MCPs | Yes | Free, no key |
Your profile (profile/me.md) and verified facts (profile/facts.md) |
Yes | Built for you in an interview, from your resume, GitHub and portfolio |
| Firecrawl MCP | Optional | Free plan, your own key |
| Gmail drafts (Google Workspace MCP) | Optional | Free, your own Google Cloud OAuth client. Drafts only, it can't send |
| Daily scheduling | Optional | schedule/run.sh (cron) or schedule/run.ps1 (Windows Task Scheduler) |
- Never sends anything. Emails become drafts. You send every email and DM by hand.
- No invented facts. Claims about you come only from
profile/facts.md. Claims about companies have a source URL. Unverified things are labelled. - Public reads only on X and LinkedIn. No logins, cookies or browser automation there, and no automated likes, follows or DMs.
- Your data stays local. Your profile, leads, briefs and resumes are gitignored, so forking this repo publicly never leaks your job hunt.
- Free tools first, one call at a time, and it backs off on rate limits.
AGENTS.md the operating manual every agent reads
SETUP.md first-run setup, written for an agent to follow
prompts/ the workflows (any agent: "read prompts/<name>.md and do it")
playbook/ discovery strategy and outreach rules
profile/ templates for your profile and verified facts
config/ MCP config for Claude Code, OpenCode, Codex, Gemini CLI, Pi
schedule/ headless daily runs
tools/ tiny free helpers (keyless search, read a public X post)
examples/ what a sweep output looks like (fictional)
Slash commands: /funding-sweep, /x-sweep, /research-company, /follow-ups in Claude Code, OpenCode, Gemini CLI and Pi.
- A sweep is a long, multi-tool run. Small free models can skip rules or stumble on tools. If results look sloppy, use a stronger model for the sweeps.
- Free search endpoints have rate limits, and some days nothing fits you. That's normal.
- The agent finds and drafts. Getting hired is still on you: proof of work, good follow-ups, and showing up well on calls.
MIT