Connect Lamr to an AI assistant that can drive a real, signed-in browser — Claude with the Chrome extension or computer use, for one — and ask it to apply. It reads the employer's live form, fills exactly what it can verify from your resume and profile, and leaves everything else for you. Nothing goes out without a real confirmation.
A real Greenhouse-style application form, resolved the way Lamr actually resolves one: name, email, and work authorization come straight from your saved profile — verified, not guessed. The optional demographic question is never touched. Once you're ready, your agent confirms and Lamr logs the outcome.
"Alex Rivera" is a placeholder candidate — not a real Lamr user. The gender field stays on its placeholder forever; it is never defaulted to whatever option happens to be listed first.
Lamr has no browser of its own — it never has, and per docs/mcp-plan.md a server-side submitter isn't viable against Cloudflare, DataDome, or a JS-rendered Workday form anyway. What changed is that Lamr now knowingly equips an agent that already controls one.
Your agent opens the employer's application in your own signed-in browser and hands Lamr the exact fields it found on the page — get_application_form_plan.
Lamr resolves each field against your saved profile, resume, and tailored documents — a verified value, or no value at all if it can't be sure. The agent's resume/cover-letter PDFs come as short-lived signed links via get_application_documents.
Your agent fills only what it was given, leaves every flagged field alone, and stops entirely for a bot challenge. record_application_outcome is the one source of truth — a job is marked applied only on a confirmed submission.
So the resolver refuses far more eagerly than it guesses. This is the same logic that runs whether the agent is filling one field or fifty — it isn't a suggestion the model can talk itself out of, there is simply nothing returned to fill.
Race, gender, veteran status, disability — never answered, ever, regardless of what the form defaults to.
"Authorized to work without sponsorship" needs two profile facts to agree — reading either one alone is a guess.
A free-text answer that doesn't map to anything the form actually offers becomes a flag, never a nearest-match guess.
No source in your resume or profile means no value — not a plausible-sounding fallback.
Auto-apply rides the exact connector described on the Lamr MCP page — no separate install, no second account link. It only needs an assistant that can drive a real, already-signed-in browser: Claude with the Chrome extension or computer use is the most capable path today. A cloud-hosted browser with no session of yours hits the same Cloudflare or DataDome walls a script would — Lamr can still draft and track the application there, just not click submit.
Worth reading before you rely on this. A browser extension can enforce a refusal as code that can't be argued out of; here, the rule lives as instructions a capable agent follows. The structural backstop is that a refused field carries no value to begin with — there's nothing to fill even if a model were inclined to.
No. Lamr has no browser of its own. It equips an AI agent that already controls a real, signed-in browser — Claude with the Chrome extension or computer use, for example — with the exact values to fill and an explicit list of what to leave alone. The agent does the clicking, in your own browser session, under your own account.
Demographic and EEO questions (race, gender, veteran or disability status) are never answered. A compound question like "authorized to work without sponsorship" is only answered when both underlying profile facts agree, not by reading one and guessing the other. A free-text answer that doesn't map to any option the form actually offers becomes a flag, not a nearest-match guess. If a field has no source in your profile or resume, it gets no value — there's nothing for the agent to fill even if it wanted to.
The agent stops and hands control back to you. A CAPTCHA is never solved or bypassed — getting around one would also void the application under most employers' own terms.
The agent reports the outcome back to Lamr, which is the one source of truth: a job is only marked applied on a confirmed submission. An interrupted or unclear attempt is recorded as unconfirmed instead, so you always know to go check rather than trusting a false positive.
Any MCP-connected assistant that can drive a real, already-signed-in browser — Claude with the Chrome extension or computer use is the most capable path today. A cloud-hosted browser with no session of yours hits the same bot defenses (Cloudflare, DataDome) a script would; Lamr can still draft and track the application there, just not click submit.
Unresolved, honestly. Greenhouse, Lever, Workday, and iCIMS generally prohibit automated submission in their terms. Running inside your own agent session under your own account reduces the technical exposure of a scraper or bot farm, but it does not resolve the contractual question — that risk is yours to weigh.
Filling the form is the last step. Everything below it — tailoring, tracking, screening — still runs through the same handlers, credits, and truth rules.
Private alpha
Leave your email and you'll get an invite the moment the doors open — one email, no spam.