AppliedIn · the agentic workflow for job applications

Open-source agent graph

AppliedIn<br>The agentic workflow for job applications.

Discover relevant roles, tailor a truthful resume, approve every change,<br>and let the graph complete the application in your browser.

Open source<br>Optional human-in-the-loop

Explore the repository<br>Try the live demo

Start with one job URL. Review the tailored resume before browser<br>automation can act in your name.

Human-in-the-loop available before browser action

Graph engineering

What it solves<br>Step ownership<br>End-to-end graph<br>Agentic patterns<br>The product

Pipeline board<br>One application run<br>Resume tailoring<br>Local tools and accounts<br>Safety

Guardrails<br>Optional review gates<br>Run it

Get started<br>What you need<br>Before auto-submit

You decide where the graph pauses. Use gated mode to<br>approve each application, or allow strong matches to continue. Missing<br>facts, logins, and uncertain outcomes still return to you.

Each step has a clear owner

AppliedIn assigns each part of an application to the actor best suited<br>for it. Code handles repeatable rules. Focused agents handle judgment and<br>changing websites. You control identity, missing facts, and submission.

Code handles the facts. Focused agents handle uncertainty. You stay in<br>control when the workflow acts in your name.

From job discovery to verified application

Every job follows explicit states. Tools discover and filter it, focused<br>agents score and tailor it, optional review gates can pause it, and the<br>browser agent verifies the final outcome.

Discover → Score → Tailor ↔ Critique → Approve → Apply → Verify.<br>Full size: light or<br>dark.

Agentic patterns behind the workflow

Sequential orchestration

Explicit stages keep the business process inspectable and make<br>recovery possible when a long-running step fails.

Structured output

The scorer returns a typed 0–10 MatchScore instead of prose that<br>downstream code has to interpret.

Review and critique

The tailor writes, tools validate, and the critic gets no more than<br>two passes to request one focused revision.

Human-in-the-loop

Approval, unknown facts, logins, and CAPTCHAs pause the workflow and<br>return control to the owner.

Tool routing

Structured feeds handle known job boards. Browser-capable discovery<br>is used only when the deterministic path cannot resolve a page.

Bounded browser autonomy

The browser observes, acts, and verifies, but cannot invent missing<br>answers or retry an uncertain submission blindly.

See every job move through the graph

Jobs move through explicit states on a local board. A role can be found,<br>tailored, waiting for you, submitting, applied, failed, or uncertain.

The real dashboard running on sample data.<br>Open the interactive demo.

From one job URL to a completed application

01

Find and score

Reads the boards you follow, removes obvious mismatches, deduplicates<br>roles, and stops low scores before expensive work begins.

02

Tailor and validate

Rewords existing resume content toward the role while code preserves<br>every bullet and anchored fact and produces a valid PDF.

03

Approve and apply

Shows what changed, waits for approval, fills the employer form from<br>approved facts, and verifies the final confirmation.

Tailor the resume without changing the facts

The tailor can adapt wording and emphasis for a role. Validation keeps<br>every source fact anchored, produces a real PDF, and limits the critic<br>to one focused revision across no more than two passes.

The agent adapts the resume for relevance. Code protects the source<br>facts and produces the final PDF.

Choose where the workflow pauses

In gated mode, you review the tailored resume before the browser acts.<br>In automatic mode, strong matches can continue without that pause.<br>Missing information, logins, CAPTCHAs, and uncertain outcomes always<br>return to you instead of becoming invented answers.

Review can be optional. Confirmation, human questions, and uncertain<br>outcomes always remain separate paths.

The same boundaries stay visible on the product board.

Optional approvals, missing information, and browser handoffs use the<br>same durable gate mechanism.

What runs locally, and what requires an account

The workflow and application history stay on your machine. Core graph<br>reasoning uses your OpenAI key. Automatic form submission uses Claude<br>Code and its Chrome integration in the browser session you control.

Core graph

your OpenAI key

Form submission

your Claude subscription

Files and history

your local machine

Safety rules the model cannot bypass

A boundary written only in a prompt is a boundary nobody is enforcing.<br>These rules also live at the tool and state layers.

Truthful resume

Every bullet and anchored fact is checked before a tailored copy is saved.

No duplicate application

A job already marked applied or applied manually is refused before submission.

No invented answer

An unknown required field becomes a human question instead of a guess.

No unproven success

Without positive...