SAYBOOKS · jobhunt

Your job hunt, on the record.

A free tracker for people running their search with an AI agent. Your Claude connects as your delegate — and works under rules: it never invents a date or a name, the duplicate guard catches the same req retitled by a fourth agency, and every action lands in an audit trail you can read.

Start free with Google → See it working first

The demo is a live sandbox with a seeded hunt — no account, poke anything. Sign in later to keep a space of your own. No credit card, no trial clock; export or delete everything, any time.

You: Log this one — "Senior Operations Manager, Lakeshore Foods", came through a TalentBridge recruiter. JD pasted below.

Claude: Logged. One thing before you apply: ⚠ this looks like the same role you already applied to directly at Lakeshore two weeks ago — same employer, near-identical description under a different title. A second submission through an agency can get you disqualified. Want me to mark it a duplicate and skip it?

What the rules catch

The double-submission guard

Agencies retitle the same req constantly. The guard matches on req id, URL, employer — and the job description body itself, reordered templates included. Warnings are surfaced for you to weigh, never silently resolved.

Never an invented fact

An application date that was never recorded stays "unknown" — the agent is refused if it tries to guess. When the true date surfaces, a correction records it with its source.

Ghosted is a status, not a deletion

Silence is an outcome, and which companies go quiet is pattern knowledge. Terminal states stay on the books, one click away.

One open next action

Per pursuit, one thing to do next, with a due date. The Due page is your morning list; superseded intentions are kept, not erased.

How it works with Claude — three steps

Sign in above.

You get your own private space for the hunt. Nothing to install.

Add Saybooks to Claude as a connector.

In the Claude app: Customize → Connectors → Add (or open claude.ai/customize/connectors), paste the address below, Continue. Claude sends you to sign in with Google and asks which books to open and what the agent may do — pick your hunt, leave controller, and you're connected. Revoke it any time under Share this space….

Name

hunt

Remote MCP server URL

https://saybooks.io/mcp/hunt

Open a new session with the connector on, and paste this — with a job description under it.

You're my job-hunt assistant, connected to my Saybooks hunt space through the "hunt" connector. Call core_schema first to learn the rules. Then log the posting below: evaluate the fit, run the duplicate check, and give me one next action with a due date.

From then on, just talk: "the recruiter called, interview Thursday" · "TCS ghosted me" · "what's due today?" Everything it records shows up in your browser too — same data, same rules, one audit trail. Works with any MCP-capable agent (Claude Code: claude mcp add hunt -t http https://saybooks.io/mcp/hunt, then /mcp to sign in).

Your data, actually yours

Export — one click downloads every table as JSON, audit log included.
Delete — one button permanently removes your space, its database, and every access key.
Self-host — the whole thing is open source (AGPL): github.com/alviso/saybooks. Run it on your laptop if you'd rather.
No LinkedIn integration, by design — nothing here touches your account or scrapes anything. You and your agent record what you know.

Built on Saybooks — the open-source ERP where AI agents and humans share one set of rules and one audit trail. The jobhunt module is the proof that the machine generalizes: same registry, same guards, pointed at a personal domain. The founder runs his own hunt on it.