What’s new

Find the keywords whose page one you can beat, and the ad patterns a whole category keeps paying for.

• Keyword research: upload page one for the keywords you're weighing. Jev asks five questions of every result (answers the exact search, forum thread, thin page, out of date, strong site) and shows which keywords have three or more beatable results, which to go after first, and why.
• Results Jev was split on go to a needs-a-human list with the probability, so you check a handful by hand.
• Ad research compares brands: the category playbook lists the patterns two or more brands each keep running past your threshold, and a brand table shows who keeps ads running longest and what they lean on.
• The ad labeler now reads awareness stage (unaware to most aware), and labels that were a coin flip between two options are flagged for a person.
• One new judge, Search result beatability. 76 in all.

250,000 judgments a month for $99, from the billing page.

• Buy Growth from the billing page, or move up from Pro with credit for the unused part of your Pro month. The exact charge shows before you confirm.
• Moving down from Growth to Pro takes effect at your next renewal, and you can keep Growth until then with one click.
• When a Pro account runs out of judgments, the message now points to Growth.

From the 70 builds added to shipwithjev.com since September: new judges, new tools, and better questions everywhere.

• Audience test: play a speech, toast, pitch or script to a room of 200 seats, line by line, and see what to rework first.
• MCP server: Claude Code, Cursor, Codex and other MCP clients can find and run judges, flows and ranking as tools.
• Line by line: run any judge on every line of a text with /api/v1/judges/{slug}/lines.
• Spec check in the judge builder: catches the question designs Jev answers confidently wrong, like a pick-one with no "none of these" option, plus an optional review by Jev.
• Should this be a judge? Paste an LLM prompt from your code and see whether a judge can take it over, and which one.
• Nine new judges: package risk, vulnerable code, bot intent, claim vs evidence, memory gate, audience reaction, draft sentence check, judge question review and should-this-be-a-judge. 75 in all.
• The prompt-injection guard takes what your assistant is for and catches topic hijacks; the tool-call firewall catches downloaded scripts and broken limits, and no longer holds up actions the task asked for.
• Eight pick-one questions across the gallery gained a "none of these" option.

Judges that run an agent team's handoffs, and a way to see which ads in a category keep running.

• Flows: chain judges into the gates an agent team runs on (worth reading, which desk, does the claim hold, new or known). Each verdict sends the item to the next gate, an agent, a delivery, storage, an escalation or the bin.
• A live control room for every flow: the graph with its traffic, the run log, each gate's pass rate, items per hour and what Jev cost.
• Four ready-made flows (Research desk, Support inbox, Agent tool-call gate, Content pipeline), callable at /api/v1/flows/{slug}/run.
• Ad research: load a category's ads with how long each has been running. Jev labels format, hook, angle, offer, proof, objection and call to action, shows which patterns survive, and scores briefs against the survivors.
• Four new judges: Ad anatomy labeler, Ad brief scorer, Research desk router and Novelty check. 66 in all.
• Batches can carry extra columns that aren't judged, like days live, and the CSV export includes them.

Fewer ways for a new account to get stuck, and a plan for every volume.

• Didn't get the email? Resend it from the sign-up and sign-in screens, with tips for finding it.
• Growth plan listed: 250,000 judgments a month for $99. Self-serve checkout opens soon.
• Scale now has a Talk to us button that carries the volume you picked on the slider.
• Share cards, a sitemap and canonical links for every page and all 62 judges.
• New About and Changelog pages.

Evally's own visual language: decisions you can watch happen.

• Live decision fields: 0s and 1s that resolve into verdicts as a judge sweeps past.
• A "Questions in. A verdict out." explainer that replays real recorded runs step by step.
• Every probability drawn as a bit meter, with the line it has to clear.
• A ticker of all 62 ready-made judges and their verdicts.

From one built-in check to a judge for every AI decision.

• 62 ready-made judges in 12 categories, each runnable live on its own page: guardrails, hallucinations, agent tool calls, moderation, sales, support, hiring, compliance and more.
• Judge builder: typed questions, rules and outcomes, versioned on every save. API calls can pin a version.
• Batches: upload CSV or JSON Lines, watch progress live, export the results and measure agreement with an expected column.
• A review queue for verdicts that need a person, with confirm or correct.
• Rank and decide APIs.

Evally opens to the public.

• Hobby stays free with 500 judgments a month. Pro gives 50,000 a month for $29.
• Checkout, invoices and sales tax handled by Dodo Payments as merchant of record.
• Sign in with a password or a magic link, and reset a forgotten password.
• Terms, privacy and refund policies.

Everything you need to call Evally from your own code.

• Real accounts, with a dashboard of your usage and recent judgments.
• API keys are stored hashed and shown once. Revoke one and it stops working at once.
• Every judgment counts against your monthly allowance, with a clear error when it runs out.
• Judged text isn't stored by default: run logs keep the verdict, timing and token counts.

Where it started: one call that grades an AI answer.

• One Jev call scores an answer for grounding, relevance, safety and quality, each with a calibrated probability.
• A pass, flag or fail verdict on top, so code can branch on it.
• POST /api/judge, a playground in the browser, and a quickstart for cURL, JavaScript and Python.