Agent examples
Each example is a complete agent you can create in your account and adapt. Pick one by use case, or search by what you want it to do.
Featured
Paste a prospect's website address and get three short, personalized outreach emails that reference what the company actually says about itself, each with a different opening angle, for you to edit and send.
Starts: when you run it
Upload a video of yourself giving a presentation, pitch or talk and get coaching feedback on pace, clarity, structure, filler words, body language and slides, with timestamped examples and three things to practise next.
Starts: when you run it
Upload one or more reference images with an instruction, and get three different takes on it, each generated from your references.
Starts: when you run it
Paste the timeline, chat excerpts and log lines from an outage and get a blameless postmortem draft: summary, impact, timeline, contributing factors, and follow-up actions for your team to review.
Starts: when you run it
Checks an RSS feed every day at 07:00 UTC, keeps the items about your topics, summarizes each one, and emails you a digest. Sends nothing on days with no matches.
Starts: on a schedule
Watches an uploaded video and its caption and decides allow, review or block against your rules, listing the moments that raised concern. Borderline videos go to a person to approve or reject.
Starts: when you run it
Paste a customer message and the reply your chatbot or AI agent wrote, and get it scored from 1 to 5 for accuracy, tone and policy compliance against the rules in the prompt, with the specific problems found. Replies that score low are sent to a person to mark as acceptable or needing a fix.
Starts: when you run it
Upload a recording of a sales or support call and get CRM-ready fields as JSON (contact, company, call type, summary, needs, objections, next step, follow-up date) plus the customer's sentiment, sent to a webhook you choose. Accepts MP3, WAV, AAC, OGG and FLAC recordings.
Starts: when you run it
Forward customer replies on support tickets to the agent. Replies that only say thanks or confirm the issue is fixed are sent to your close-ticket webhook; anything that asks for more, reports a new problem or is unclear goes to your support queue webhook instead.
Starts: when an email arrives
Scores each customer message for frustration and for signs the customer may cancel or leave, such as competitor mentions or cancellation threats. Messages above your threshold are posted to a webhook you choose, with the reasons and a suggested next step.
Starts: when you run it
Paste a customer question and its answer and get a short video clip that shows the idea visually, for your help center or social channels. The clip has no on-screen text, so pair it with the written answer.
Starts: when you run it
Forward your customer inbox to the agent. When a message mentions an injury, an allergic reaction or a product safety hazard, you get an email straight away with the category and the customer's own words quoted, so a person can follow up and decide whether it must be reported. It flags messages; it does not assess them.
Starts: when an email arrives
Answers questions from your documents. Rewrites the question for search, retrieves the best passages from a knowledge base, and replies with linked citations.
Starts: when you run it
Upload up to 10 voicemails (MP3, WAV, AAC, OGG or FLAC). The agent listens to each one and builds a callback list with the caller's name and number, the kind of matter, any deadline or court date they mention, and an urgency label, most urgent first. It is a draft for your intake staff to check against the recordings; it gives no legal assessment.
Starts: when you run it
Upload a photo, optionally with a question about it, and get a spoken description written for listening: what the scene is, who and what is in it, and any readable text. You get the description as text and as an audio file.
Starts: when you run it
Give people an address to email privacy requests to. Each email is sorted into access, deletion, correction, opt-out or not a privacy request, the requester's details are pulled out, and a triage note is emailed to you so a person can verify identity and respond.
Starts: when an email arrives
Drafts a public reply to a customer review, matching the tone to the star rating: thanks for happy customers, an apology and a next step for unhappy ones. Replies to 1- and 2-star reviews wait for your approval; the rest go straight to your publishing webhook.
Starts: when you run it
Upload or paste a CSV of questions with the answers you expect. For each question the agent retrieves passages from your knowledge base, answers from them, and has a judge score the answer for correctness against your expected answer and for faithfulness to the passages. You get a summary of the failures to fix in your documents or prompts.
Starts: when you run it
Pulls the order number, amount and reason out of a customer's refund or return request and checks it against the refund policy you write into the prompt. The check uses only what the customer wrote, with no order lookup. A request is refunded automatically only when the model judges it within policy and the amount it states is more than 0 and at most 50; everything else, including any request that states no amount, waits for a person to approve or deny it.
Starts: when you run it
When a prospect emails your sales address, drafts a reply from the product and pricing facts you write into its instructions, asks you to approve it, and sends the approved reply to the sender. Nothing is sent if you decline or don't answer within a day.
Starts: when an email arrives
Upload a recorded support call and get it scored against the quality checklist in the prompt (greeting, verification, understanding the issue, resolution, tone, closing), with evidence and coaching notes for each item. When the scorer is unsure, a supervisor is asked to confirm the score or rescore the call by hand. Accepts MP3, WAV, AAC, OGG and FLAC recordings.
Starts: when you run it
A streaming support chatbot that answers from your documentation and remembers each conversation, so follow-up questions keep their context.
Starts: when you run it
Customers email the agent's inbox; it looks up the answer in your help documentation and drafts a reply. You approve each draft before it goes back to the customer, quoted under their original message.
Starts: when an email arrives
Forward your support inbox to the agent. It labels each email with a category, urgency and customer sentiment, posts the labels to your helpdesk webhook, and emails you straight away when an email is urgent.
Starts: when an email arrives
Paste a customer message. The agent decides whether it is about billing, a technical problem, a sales question or something else, and hands it to a specialist with its own instructions for that kind of question, which drafts the reply for you to send.
Starts: when you run it
Paste a long ticket thread or chat transcript and get a short handoff note for the next agent or shift: the customer's problem, what has been tried, what was promised, how the customer feels, and the next step.
Starts: when you run it
Send each new ticket from your helpdesk or contact form through the API. The agent decides whether it belongs to billing, technical support or account management, sets a priority, and posts it to that team's webhook; anything else goes to a general queue.
Starts: when you run it
Tells you whether to block, redact or allow a message: returns a blocked notice, the redacted message, or the message unchanged.
Starts: when you run it
Listens to a voice message and returns a transcript, the exact phrases that break your rules, and a verdict: allow or block, with the reason. Accepts MP3, WAV, AAC, OGG and FLAC recordings.
Starts: when you run it
Upload a voice message and name the language you want; get back the original transcript, the translation, and the translation read aloud as an audio file. With no language named, it translates into Spanish.
Starts: when you run it
Describe the deal (parties, scope, price, term and anything unusual) and a first draft is assembled from the contract templates and approved clauses you upload. Anything the terms don't settle is marked [TO CONFIRM] instead of filled in, and a list of deviations from your templates is added. A legal reviewer then records whether it is ready after edits or needs redrafting; the choice is recorded in the run and nothing is sent. It is a draft for a lawyer to check, not legal advice.
Starts: when you run it
Upload a recording of a sales or support call and get CRM-ready fields as JSON (contact, company, call type, summary, needs, objections, next step, follow-up date) plus the customer's sentiment, sent to a webhook you choose. Accepts MP3, WAV, AAC, OGG and FLAC recordings.
Starts: when you run it
Paste a prospect's website address and get three short, personalized outreach emails that reference what the company actually says about itself, each with a different opening angle, for you to edit and send.
Starts: when you run it
Paste a list of company domains or upload a CSV. The agent reads each company's home page and records its industry, what it sells, size signals and how well it fits your ideal customer profile, with the reason. Each record is posted as JSON to your webhook, and you get a summary table at the end. A company whose website cannot be processed is listed as not processed instead of stopping the run.
Starts: when you run it
Every Monday at 08:00 UTC, reads the competitor pages you list (pricing, home or product pages), summarizes each one's plans, prices and headline claims, and posts a side-by-side snapshot with talking points to your webhook.
Starts: on a schedule
Reads a contact-form or demo-request submission, pulls out the contact details, scores how well the lead fits your ideal customer, and sends leads scoring 70 or more to your sales webhook. Lower scores are shown with the reason instead.
Starts: when you run it
Upload a video of yourself giving a presentation, pitch or talk and get coaching feedback on pace, clarity, structure, filler words, body language and slides, with timestamped examples and three things to practise next.
Starts: when you run it
Upload an RFP or security questionnaire. The agent pulls out each question, searches your past answers and product documents for it, and drafts an answer with its source, or marks it "needs input" when your documents don't cover it. A reviewer gets the full draft to sign off before you use it.
Starts: when you run it
Paste a call transcript or your rough notes and get the deal fields (stage, budget, decision makers, pain points, next step and date) as JSON, shown to you and posted to a webhook you choose (for example an automation tool that updates your CRM).
Starts: when you run it
Give it a company name or website and get a one-page brief before your call: what the company does, recent news, likely priorities, and questions to ask, with a link for every fact.
Starts: when you run it
When a prospect emails your sales address, drafts a reply from the product and pricing facts you write into its instructions, asks you to approve it, and sends the approved reply to the sender. Nothing is sent if you decline or don't answer within a day.
Starts: when an email arrives
Paste a client's requirements, RFP excerpt or discovery notes and get a structured first draft of a proposal: their goals in their words, your proposed approach, scope, timeline and open questions, with pricing left as placeholders for you to fill in and check.
Starts: when you run it
Upload a customer's security questionnaire as a PDF or Word file. Each question is answered from your security policies and documents in a knowledge base, with the source document linked, or marked "Needs input" when your documents do not cover it. The result is a draft for a person to check before it goes to the customer.
Starts: when you run it
Describe your product, audience and offer, and get three distinct ad variants to test against each other, each with a headline, body copy, a call to action and a generated image.
Starts: when you run it
Paste a product announcement, press release or blog post and get a post written for each network's style and length: LinkedIn, X, Instagram, Facebook and Threads.
Starts: when you run it
Paste a blog post and get a header illustration in one fixed house style. Each image is checked against the style and the post; one that falls short is regenerated, up to three images in total, and you see the first that passes. If none passes, the run ends without a result; the run shows the last image and its score, earlier images are not kept, and every image is charged.
Starts: when you run it
Give it a topic or working title and it searches the web, reads the best sources, and writes an outline plus a full first draft of a blog post in your brand voice, with the sources it used linked for fact-checking.
Starts: when you run it
Checks an image against your brand guidelines. Images the check passes are marked approved without review; anything it flags or is unsure about goes to a person.
Starts: when you run it
Every day at 07:00 UTC, searches the web for your brand or product name and posts the results that mention it to your webhook, each with a one-line summary and whether the mention reads as positive, neutral or negative. Posts nothing when no result mentions it. The same page can come up on more than one day.
Starts: on a schedule
Upload a few images that show your brand's look (past ads, product shots, your style guide's sample imagery) with a campaign brief, and get four campaign image concepts generated to match that look.
Starts: when you run it
Paste a prospect's website address and get three short, personalized outreach emails that reference what the company actually says about itself, each with a different opening angle, for you to edit and send.
Starts: when you run it
Every Monday at 08:00 UTC, reads the competitor pages you list (pricing, home or product pages), summarizes each one's plans, prices and headline claims, and posts a side-by-side snapshot with talking points to your webhook.
Starts: on a schedule
Writes a draft from your brief, critiques it against your criteria, rewrites it to fix every point, then scores the rewrite. If the score is below the bar it starts again with a fresh draft, up to three attempts in total, and shows the first rewrite that passes. If none of the three passes, the run ends without a result; the run shows the last attempt and its score, earlier attempts are not kept, and every attempt is charged.
Starts: when you run it
Describe your startup and get a first-draft go-to-market plan. A market analyst, a positioning strategist and a copywriter each work on the web search results at the same time, then an editor combines their work into one plan for you to check and adapt.
Starts: when you run it
Type a seed topic. The agent searches the web, reads the top-ranking pages, groups related keyword ideas by search intent, and writes a content brief for a writer: angle, headings, questions to answer and the gaps the current results leave. Keyword ideas come from what the pages cover; it has no search volume or difficulty data.
Starts: when you run it
Paste a page URL and the search phrase you want it to rank for. The agent reads the page and gives you a prioritised list of fixes for the title, meta description, headings, body copy and calls to action, each with the current text and a suggested rewrite. It checks only what is on the page, not rankings, backlinks or speed.
Starts: when you run it
Upload a product or property listing photo and get scores from 1 to 5 for sharpness, lighting, framing and background, a keep or retake verdict, and specific tips for a better retake.
Starts: when you run it
Paste the links you want to share this week, with optional notes, and get a ready-to-edit email newsletter: a subject line, an intro, and a short summary of each article with its link.
Starts: when you run it
Paste your episode notes and sponsor details and get three scripts, an episode intro, an outro and a sponsor ad read, each recorded as an audio file you can drop into your edit. The ad read only uses claims from the sponsor details you give.
Starts: when you run it
Upload product photos with your specs or supplier notes and get a ready-to-edit listing: a product title, five benefit-led bullet points, a description paragraph and search keywords, written to stick to your specs.
Starts: when you run it
Describe what you are launching and get a matching set: a social announcement post, a launch email with subject line, a header image and a short spoken audio teaser, all written from the same brief.
Starts: when you run it
Upload a photo of your product and optionally describe the look you want. You get three new images of the same product: a clean studio shot on a plain background and two lifestyle scenes for your store or ads.
Starts: when you run it
Paste a product listing and get it localized for each market on your list (German, French, Spanish and Japanese to start): wording that reads naturally to local shoppers and metric or imperial equivalents next to the original measurements, with prices, brand names, model numbers and SKUs left untouched.
Starts: when you run it
Drafts a public reply to a customer review, matching the tone to the star rating: thanks for happy customers, an apology and a next step for unhappy ones. Replies to 1- and 2-star reviews wait for your approval; the rest go straight to your publishing webhook.
Starts: when you run it
Describe a product or event and get a short promotional video clip, with the shot written for you first.
Starts: when you run it
Upload photos of a property and type its facts (bedrooms, size, price, features, neighborhood) to get a listing description written from them. A second pass checks the wording for phrases that describe who should live there rather than the property, such as references to family status, religion, age or disability. If anything is flagged, you are asked to review it before using the listing. The check is a wording screen, not legal advice.
Starts: when you run it
Paste a list of page URLs and get, for each page, a search-friendly title aimed at about 60 characters and a meta description aimed at about 155 characters written from what the page actually says, ready to copy into your CMS.
Starts: when you run it
Upload a hand-drawn sketch, whiteboard drawing or wireframe and say what it is; get back a polished render that keeps your layout, plus the brief used to make it.
Starts: when you run it
Drafts a social media post and an illustration from your brief, waits for a person to approve the post text, then sends the approved post and its image to your publishing webhook. The approval request shows the text only.
Starts: when you run it
Every Monday at 08:00 UTC, searches the web for recent news and discussion in your niche and sends up to ten content ideas to your webhook, each with a hook, a suggested format and the link that inspired it. Set your niche in the trigger's input.
Starts: on a schedule
Paste a supplier's product line, spreadsheet row or spec sheet text and get standard catalog fields back as validated JSON: brand, product type, color, size, material, dimensions in centimetres, weight in kilograms and barcode. The JSON is posted to a webhook you choose (for example an automation tool that updates your catalog).
Starts: when you run it
Every day at 09:00 UTC, reads a product or pricing page and posts to your webhook on each day the price of the item you name is below your target. It does not remember earlier prices, so it posts every day the price stays below target. A price it cannot find on the page also posts.
Starts: on a schedule
Upload a product or property listing photo and get scores from 1 to 5 for sharpness, lighting, framing and background, a keep or retake verdict, and specific tips for a better retake.
Starts: when you run it
Reads a seller's listing text and photos and checks them for prohibited items, scam signals and misleading claims, then approves, rejects or sends the listing to a person to review.
Starts: when you run it
Upload product photos with your specs or supplier notes and get a ready-to-edit listing: a product title, five benefit-led bullet points, a description paragraph and search keywords, written to stick to your specs.
Starts: when you run it
Upload a photo of your product and optionally describe the look you want. You get three new images of the same product: a clean studio shot on a plain background and two lifestyle scenes for your store or ads.
Starts: when you run it
Paste a product listing and get it localized for each market on your list (German, French, Spanish and Japanese to start): wording that reads naturally to local shoppers and metric or imperial equivalents next to the original measurements, with prices, brand names, model numbers and SKUs left untouched.
Starts: when you run it
Paste a batch of customer reviews for a product and see what shoppers love, what they complain about, the recurring themes with how often each comes up, short quotes as evidence, and fixes worth considering.
Starts: when you run it
Drafts a public reply to a customer review, matching the tone to the star rating: thanks for happy customers, an apology and a next step for unhappy ones. Replies to 1- and 2-star reviews wait for your approval; the rest go straight to your publishing webhook.
Starts: when you run it
Pulls the order number, amount and reason out of a customer's refund or return request and checks it against the refund policy you write into the prompt. The check uses only what the customer wrote, with no order lookup. A request is refunded automatically only when the model judges it within policy and the amount it states is more than 0 and at most 50; everything else, including any request that states no amount, waits for a person to approve or deny it.
Starts: when you run it
When a job candidate emails your recruiting address with a question about the process, benefits, interview format or timeline, a reply is drafted from the recruiting FAQ documents you upload. A recruiter approves the draft before it is sent to the candidate; nothing is sent if they decline or don't answer within a day. Questions the FAQ does not cover are left for the recruiter to answer.
Starts: when an email arrives
Upload exit interview notes or an employee survey export (CSV, PDF, Word or text) and get the recurring themes for each team and across the company, with how often each came up and a paraphrased example. The instructions keep the summary at the aggregate level: no names, no quotes that could identify a person, and teams with fewer than five responses are folded into one group. It is a summary for HR to read, not an assessment of any individual or manager.
Starts: when you run it
Lets employees ask questions about leave, benefits, expenses and other policies and get answers drawn only from your employee handbook and policy documents, with links to the passages used. Says so when the handbook does not cover a question.
Starts: when you run it
Paste a job description and get a structured interview plan: questions grouped by skill, what a strong answer looks like, and a 1-4 scoring rubric so every interviewer rates candidates on the same criteria.
Starts: when you run it
Turn a handful of bullet points about a role into a complete, inclusive job posting draft with responsibilities, requirements and nice-to-haves, ready for you to edit.
Starts: when you run it
Upload your CV and describe the role, location and anything else you want. The agent searches the web for current openings, reads up to five postings, and for each one shows how your CV matches its requirements, what's missing, and a tailored cover letter draft that uses only experience from your CV.
Starts: when you run it
Give a new hire's role, team and start date and get a first-90-days onboarding checklist split into before day one, week one, the first month and the first three months, with an owner for each task.
Starts: when you run it
Upload a contract, terms of service or internal policy and get a draft plain-language version, section by section next to the original wording, with terms that need a careful read pointed out. It is a draft for a person to check, not legal advice.
Starts: when you run it
Paste the requirements for a role and upload a résumé. Get a requirement-by-requirement summary of what the résumé shows, with quotes, for a recruiter or hiring manager to read before they decide. It does not score, rank or reject candidates.
Starts: when you run it
Paste an application or upload a CV and up to six employers it names (the most recent first) are searched on the web. You get a list of things worth checking (employers with no web presence, dates that overlap or don't add up, job titles that don't match the company) and a low, medium or high level of concern. It never rejects anyone: a recruiter reads the notes and decides what to verify.
Starts: when you run it
Employees email their vacation or leave requests to an address you give them. The dates and type of leave are pulled out and a manager or admin is asked to approve or reject. An approved request is posted to your HR system's webhook and the employee gets a reply saying it was approved; a rejected one gets a reply saying so. If nobody answers within three days, nothing is sent.
Starts: when an email arrives
Give customers an address for billing questions. Each email is sorted into remittance advice, invoice dispute, invoice copy request, tax form request or other, and the invoice number and amount are pulled out. Each of the four billing types is posted to its own webhook; an email sorted as other is not posted anywhere. A suggested reply is drafted for a person to check and send; nothing is sent to the customer.
Starts: when an email arrives
Upload your annual reports, each with a year metadata field, then ask something like 'Compare 2023 and 2024 on revenue drivers and risks'. The agent searches each year's reports separately, notes what they say, and writes a side-by-side summary of what changed, with citations, for a person to check.
Starts: when you run it
Upload a bank or card statement PDF and get every transaction as a row with date, description, amount and a spending category you define, as a CSV table you can paste into a spreadsheet. The model reads the rows and then a second pass adds them up again and compares the result with the statement's opening and closing balances, listing any mismatch. The arithmetic is done by the model and checked once, so treat the table as a draft to reconcile, not as your books.
Starts: when you run it
Upload photos of damage to a vehicle, property or shipped goods and get a draft damage description for an adjuster or claims handler to check: what is damaged, where, the visible extent, possible causes the photos are consistent with, and what further photos or information would help. It does not estimate costs or decide the claim.
Starts: when you run it
Submit an expense report with its receipts and have each line checked against your expense policy. Reports with possible exceptions go to a manager or finance admin, who approves them or flags them for follow-up; nothing is sent to the submitter. Reports with none are summarized straight away.
Starts: when you run it
Paste this month's figures, budget and last month or last year as a table or CSV, and get a draft management commentary: headline results, the largest variances with their amounts and percentages, and questions to answer before the report goes out. The finance team adds the reasons behind the numbers.
Starts: when you run it
Claimants email their claim and supporting files to an address you give them. The email and attachments are checked against your list of required items. If something is missing, the claimant gets a reply listing what is missing. Each email is checked on its own, so the reply asks the claimant to send the whole claim again in one email, with every document, including the missing ones. If everything is there, the claim details are posted to your claims system's webhook for a handler. It checks that items are present, not whether a claim is valid or covered.
Starts: when an email arrives
Give suppliers an address to email invoices to. Each PDF invoice is read for supplier, invoice number, dates, PO number, line items and totals, and the fields are posted as JSON to a webhook you choose (for example an automation tool that updates your accounts payable system), for someone to check before payment.
Starts: when an email arrives
Upload the documents an applicant sent (PDFs or photos) and see which items on your required-documents list are present and which are missing, with key figures such as stated income and requested amount and any figures that disagree between documents. A loan officer then records whether the file is complete or needs more documents; the choice is recorded in the run and nothing is sent to the applicant. It does not assess creditworthiness or make a lending decision.
Starts: when you run it
Every day at 12:00 UTC, searches the news for each stock or company on your watchlist, summarizes what was reported with links and a label for the tone of the coverage, then combines the summaries into one readable brief that is posted to your webhook. It covers news only, not prices, and it is a reading summary, not investment advice.
Starts: on a schedule
Upload a photo or PDF of a receipt or invoice and send its merchant, date, totals and category as JSON to a webhook you choose (for example an automation tool that updates a spreadsheet or your expense tool).
Starts: when you run it
Enter a vendor's name and get a summary of what public web sources say about it: who they are, recent news, security incidents or breaches, lawsuits or regulatory actions, and financial health signals, each with a link. A starting point for your due diligence, not a risk rating.
Starts: when you run it
Describe the deal (parties, scope, price, term and anything unusual) and a first draft is assembled from the contract templates and approved clauses you upload. Anything the terms don't settle is marked [TO CONFIRM] instead of filled in, and a list of deviations from your templates is added. A legal reviewer then records whether it is ready after edits or needs redrafting; the choice is recorded in the run and nothing is sent. It is a draft for a lawyer to check, not legal advice.
Starts: when you run it
Upload a contract and get a summary of its key terms (parties, term, renewal, payment, liability, termination, governing law) plus the clauses that differ from the positions in your playbook, each quoted, for a lawyer or contract owner to review. It is a reading aid, not legal advice.
Starts: when you run it
Forward your customer inbox to the agent. When a message mentions an injury, an allergic reaction or a product safety hazard, you get an email straight away with the category and the customer's own words quoted, so a person can follow up and decide whether it must be reported. It flags messages; it does not assess them.
Starts: when an email arrives
Upload up to 10 voicemails (MP3, WAV, AAC, OGG or FLAC). The agent listens to each one and builds a callback list with the caller's name and number, the kind of matter, any deadline or court date they mention, and an urgency label, most urgent first. It is a draft for your intake staff to check against the recordings; it gives no legal assessment.
Starts: when you run it
Upload an NDA and compare it with your standard positions. When it differs, a person on your team is asked to review the listed deviations before anyone signs; when it matches, you get the summary straight away.
Starts: when you run it
Upload a PDF and get its text back with names, contact details, ID numbers and other personal data replaced by labels like [NAME], plus a list of what was removed and where, for a person to check before sharing.
Starts: when you run it
Upload a contract, terms of service or internal policy and get a draft plain-language version, section by section next to the original wording, with terms that need a careful read pointed out. It is a draft for a person to check, not legal advice.
Starts: when you run it
Give people an address to email privacy requests to. Each email is sorted into access, deletion, correction, opt-out or not a privacy request, the requester's details are pulled out, and a triage note is emailed to you so a person can verify identity and respond.
Starts: when an email arrives
Upload photos of a property and type its facts (bedrooms, size, price, features, neighborhood) to get a listing description written from them. A second pass checks the wording for phrases that describe who should live there rather than the property, such as references to family status, religion, age or disability. If anything is flagged, you are asked to review it before using the listing. The check is a wording screen, not legal advice.
Starts: when you run it
Every Monday at 07:00 UTC, reads a regulator's or standards body's news page, summarizes the updates posted in the past week and what they may mean for your team, and posts the summary to a webhook you choose. Posts nothing in a week with no updates.
Starts: on a schedule
Enter a vendor's name and get a summary of what public web sources say about it: who they are, recent news, security incidents or breaches, lawsuits or regulatory actions, and financial health signals, each with a link. A starting point for your due diligence, not a risk rating.
Starts: when you run it
Send an employee's access request in their own words. The agent checks it against the access policy you write into the prompt (least privilege, a stated business reason and a time limit) and passes that assessment to an owner or admin as advice. Every request needs a person's approval: only requests they approve are sent to your provisioning webhook. The check reads only the request text, so the requester named in the request is not verified, and nothing looks up the person's role or current access.
Starts: when you run it
Send it an alert payload from your monitoring tool. It rates the severity, writes a one-line summary, and forwards critical alerts to your paging webhook and warnings to a second webhook you choose. Informational alerts are only shown in the run result.
Starts: when you run it
Paste an OpenAPI snippet, or an example request and response, and get readable reference documentation: what the endpoint does, its parameters and fields in a table, error responses, and a curl example.
Starts: when you run it
Upload a screenshot of something broken, add a line about what you were doing, and get a structured bug report as JSON: a title, steps to reproduce, expected and actual behaviour, error text read from the screen, the environment visible in the screenshot, and a suggested severity.
Starts: when you run it
Paste a diff or a changed function and get a review: likely bugs, security and performance risks, unclear names, and the tests worth adding, each tied to the lines it is about. A second pair of eyes, not a replacement for your team's review.
Starts: when you run it
Paste an alert or error and your repository. Using your GitHub MCP connection, the agent looks at recently merged pull requests and commits, reads what they changed, and ranks the ones most likely to be related, with its reasoning, as leads for you to check. It needs the GitHub MCP tools list_pull_requests, get_pull_request, get_pull_request_files and list_commits. Tool names are not checked against your server: the agent still runs without any it cannot find, and it says so in its answer when it could not read pull requests.
Starts: when you run it
Paste an error, stack trace or a few log lines and get a plain-language explanation of what most likely went wrong, the other possible causes, and the next debugging steps to try.
Starts: when you run it
Every morning at 08:00 UTC, looks through a repository's open GitHub issues using your GitHub MCP connection and emails you what needs attention.
Starts: on a schedule
Paste the timeline, chat excerpts and log lines from an outage and get a blameless postmortem draft: summary, impact, timeline, contributing factors, and follow-up actions for your team to review.
Starts: when you run it
Employees ask how to fix a laptop, VPN, printer or account problem and get step-by-step instructions from your own IT guides, with links to the guide used. When the guides do not cover it, the answer says so and tells them how to open a ticket.
Starts: when you run it
Employees email IT with a problem, optionally with screenshots. The agent categorizes the request, rates its urgency, sends the triaged ticket as JSON to your helpdesk webhook, and replies to the employee with self-help steps to try while they wait.
Starts: when an email arrives
Staff forward a suspicious email, with screenshots if they have them, and get a reply rating it phishing, suspicious or looks safe, with the reasons: sender and header oddities, where the links really go, and pressure tricks. The agent opens only the main site of each linked domain (for example https://example.com), never the link itself, to see what is there. This is an AI review, not a reputation or blocklist lookup, and "looks safe" is not a guarantee.
Starts: when an email arrives
Upload or paste a CSV of questions with the answers you expect. For each question the agent retrieves passages from your knowledge base, answers from them, and has a judge score the answer for correctness against your expected answer and for faithfulness to the passages. You get a summary of the failures to fix in your documents or prompts.
Starts: when you run it
Paste a list of merged pull requests, commit messages or ticket titles and get release notes grouped into new features, improvements and fixes, written for your users, with internal-only changes left out.
Starts: when you run it
Paste the notes, open tickets and log entries from a shift and get a handover for the next shift: what is still open, what needs watching, what was done, and who to contact.
Starts: when you run it
Paste your table definitions and ask a question in plain words. The agent writes a SQL query, checks that every table and column it uses exists in your schema, fixes what does not, and scores the result; if the score is below the bar it starts again, up to three attempts in total. Nothing is run against a database: you get the query to review and run yourself. If no attempt passes, the run ends without a result.
Starts: when you run it
Checks a vendor's public status page every hour and, whenever it reports an outage, degraded performance or maintenance in progress, posts a short summary to a webhook you choose. It posts again each hour the problem is still listed.
Starts: on a schedule
Paste or upload a purchase order or booking request and its details (customer, items, quantities, prices, dates, delivery address) are pulled out and shown to a person field by field. They are posted to your webhook only if that person approves. The reviewer cannot edit fields here: rejecting (or not answering within a day) sends nothing, so fix the source document and submit it again.
Starts: when you run it
Upload photos of damage to a vehicle, property or shipped goods and get a draft damage description for an adjuster or claims handler to check: what is damaged, where, the visible extent, possible causes the photos are consistent with, and what further photos or information would help. It does not estimate costs or decide the claim.
Starts: when you run it
Submit an expense report with its receipts and have each line checked against your expense policy. Reports with possible exceptions go to a manager or finance admin, who approves them or flags them for follow-up; nothing is sent to the submitter. Reports with none are summarized straight away.
Starts: when you run it
Lets employees ask questions about leave, benefits, expenses and other policies and get answers drawn only from your employee handbook and policy documents, with links to the passages used. Says so when the handbook does not cover a question.
Starts: when you run it
Paste the timeline, chat excerpts and log lines from an outage and get a blameless postmortem draft: summary, impact, timeline, contributing factors, and follow-up actions for your team to review.
Starts: when you run it
Claimants email their claim and supporting files to an address you give them. The email and attachments are checked against your list of required items. If something is missing, the claimant gets a reply listing what is missing. Each email is checked on its own, so the reply asks the claimant to send the whole claim again in one email, with every document, including the missing ones. If everything is there, the claim details are posted to your claims system's webhook for a handler. It checks that items are present, not whether a claim is valid or covered.
Starts: when an email arrives
Employees ask how to fix a laptop, VPN, printer or account problem and get step-by-step instructions from your own IT guides, with links to the guide used. When the guides do not cover it, the answer says so and tells them how to open a ticket.
Starts: when you run it
Employees email IT with a problem, optionally with screenshots. The agent categorizes the request, rates its urgency, sends the triaged ticket as JSON to your helpdesk webhook, and replies to the employee with self-help steps to try while they wait.
Starts: when an email arrives
Paste meeting notes or a transcript and get the decisions made and a table of action items with owner and due date, ready to paste into your tracker. Items with no owner are flagged instead of assigned.
Starts: when you run it
Upload a video recording of a meeting or webinar and get a short summary, the decisions made, open questions, action items with owners when they are named, and a chapter list with timestamps.
Starts: when you run it
Paste a meeting transcript and get every follow-up meeting, deadline and reminder it agreed on as calendar events in JSON: title, date, start and end time, attendees and notes, ready to import or send to your own tooling.
Starts: when you run it
Give a new hire's role, team and start date and get a first-90-days onboarding checklist split into before day one, week one, the first month and the first three months, with an owner for each task.
Starts: when you run it
Checks an uploaded profile photo: does it show one real person's face, is it free of text, logos and contact details, and is it suitable for your site. Accepts or rejects it with a reason, and asks a person when it is unsure.
Starts: when you run it
Forward supplier order confirmations, purchase orders or delivery notices by email, in the body or as a PDF, and the order number, supplier, dates, line items and total are sent as JSON to a webhook you choose (for example an automation tool that updates your inventory or a spreadsheet).
Starts: when an email arrives
Upload a screen recording of someone doing a task in an app and get a written how-to guide: what it is for, what you need first, and numbered steps naming the buttons and fields used, with the timestamp where each step happens.
Starts: when you run it
Paste the notes, open tickets and log entries from a shift and get a handover for the next shift: what is still open, what needs watching, what was done, and who to contact.
Starts: when you run it
Turn rough notes, a recorded walkthrough transcript or a chat thread about how a task is done into a clear standard operating procedure with purpose, roles, prerequisites, numbered steps and checks, for the process owner to review.
Starts: when you run it
Upload a recorded support call and get it scored against the quality checklist in the prompt (greeting, verification, understanding the issue, resolution, tone, closing), with evidence and coaching notes for each item. When the scorer is unsure, a supervisor is asked to confirm the score or rescore the call by hand. Accepts MP3, WAV, AAC, OGG and FLAC recordings.
Starts: when you run it
Employees email their vacation or leave requests to an address you give them. The dates and type of leave are pulled out and a manager or admin is asked to approve or reject. An approved request is posted to your HR system's webhook and the employee gets a reply saying it was approved; a rejected one gets a reply saying so. If nobody answers within three days, nothing is sent.
Starts: when an email arrives
Upload photos of a whiteboard or sketch and get clean meeting notes with the decisions and next steps.
Starts: when you run it
A chat tutor for a course or training program. It answers learners' questions from the course materials you upload, points to where in the materials the answer is, and remembers the conversation so follow-up questions work.
Starts: when you run it
Paste or upload a student essay and get draft feedback organized by your rubric's criteria, with strengths, suggestions and a suggested level for each, for the teacher to review before sharing.
Starts: when you run it
Name any concept and get three explanations side by side: one a child could follow, one for a high-school student, and one for someone with a background in the field, each with an example.
Starts: when you run it
Paste your study notes or a textbook section and get a deck of question-and-answer flashcards as JSON, ready to import into a flashcard app or spreadsheet.
Starts: when you run it
Give a story idea, the child's name or age, and anything the story should include. You get a six-page picture-book story and one illustration per page, with the characters described the same way on every page so they look alike.
Starts: when you run it
Type a topic, a grade level and a lesson length, and get a lesson plan with learning objectives, a timed sequence of activities, materials, differentiation ideas and an exit ticket.
Starts: when you run it
Upload a video of yourself giving a presentation, pitch or talk and get coaching feedback on pace, clarity, structure, filler words, body language and slides, with timestamped examples and three things to practise next.
Starts: when you run it
Upload lecture notes, slides exported to PDF or a reading, and get a ready-to-use quiz with multiple-choice and short-answer questions, plus a separate answer key.
Starts: when you run it
Upload a research paper as a PDF and get a structured summary: the question it asks, the method, the key findings with numbers, the limitations the authors note, and open questions.
Starts: when you run it
Paste your answer key and a student's short answers and get suggested marks per question, with partial credit and a reason for each. When any answer is borderline, the teacher is asked to confirm the marks or adjust them by hand; the marks are a suggestion for the teacher, never sent to the student.
Starts: when you run it
Upload a recording, such as a meeting, voice memo or podcast, and get a short summary back as text and as spoken audio.
Starts: when you run it
A general-purpose assistant. The simplest starting point: one model call you can give your own instructions.
Starts: when you run it
Dump everything on your mind, in any order, with your working hours and meetings, and get a prioritized plan for the day: the top three tasks, a time-blocked schedule, and what can wait.
Starts: when you run it
Name any concept and get three explanations side by side: one a child could follow, one for a high-school student, and one for someone with a background in the field, each with an example.
Starts: when you run it
Forward any email to the agent and get a reply telling you how urgent it is, what it asks of you in two sentences, and a suggested reply you can copy and edit.
Starts: when an email arrives
Upload your CV and describe the role, location and anything else you want. The agent searches the web for current openings, reads up to five postings, and for each one shows how your CV matches its requirements, what's missing, and a tailored cover letter draft that uses only experience from your CV.
Starts: when you run it
Paste meeting notes or a transcript and get the decisions made and a table of action items with owner and due date, ready to paste into your tracker. Items with no owner are flagged instead of assigned.
Starts: when you run it
Upload a video recording of a meeting or webinar and get a short summary, the decisions made, open questions, action items with owners when they are named, and a chapter list with timestamps.
Starts: when you run it
Paste a meeting transcript and get every follow-up meeting, deadline and reminder it agreed on as calendar events in JSON: title, date, start and end time, attendees and notes, ready to import or send to your own tooling.
Starts: when you run it
Upload a photo and say who the card is for, the occasion and a few personal details. You get a short card message and a card illustration based on your photo.
Starts: when you run it
Upload a video of yourself giving a presentation, pitch or talk and get coaching feedback on pace, clarity, structure, filler words, body language and slides, with timestamped examples and three things to practise next.
Starts: when you run it
Upload a contract, terms of service or internal policy and get a draft plain-language version, section by section next to the original wording, with terms that need a careful read pointed out. It is a draft for a person to check, not legal advice.
Starts: when you run it
Email a link to an article you don't have time to read and get a reply with the key points, how long the full piece is, and whether it's worth reading in full.
Starts: when an email arrives
Upload a PDF report or article and get a spoken, podcast-style narration of it as an audio file, along with the script that was read.
Starts: when you run it
Upload a small CSV and ask about it in plain words, then keep asking follow-up questions in the same conversation. The model copies the uploaded table into the conversation, so check its first answer against your file; follow-up questions can refer to the table until it falls outside the roughly 50 most recent exchanges. It suits small files: the model reads the rows and does the arithmetic itself, so check totals that matter.
Starts: when you run it
Paste a long ticket thread or chat transcript and get a short handoff note for the next agent or shift: the customer's problem, what has been tried, what was promised, how the customer feels, and the next step.
Starts: when you run it
Email a voice memo and get a reply with a written summary, the action items, and a short spoken recap.
Starts: when an email arrives
Upload a voice message and name the language you want; get back the original transcript, the translation, and the translation read aloud as an audio file. With no language named, it translates into Spanish.
Starts: when you run it
Paste a web address and get a risk level (high, medium or low) with the reasons. The agent searches the web for scam reports and reviews of the domain and tries to read the site's main page (https:// plus the domain), then weighs what it found. When the site cannot be loaded, the rating rests on the search results alone. This is an AI review of public information, not a reputation database lookup, so a low rating is not a guarantee.
Starts: when you run it
Upload photos of a whiteboard or sketch and get clean meeting notes with the decisions and next steps.
Starts: when you run it
Upload your annual reports, each with a year metadata field, then ask something like 'Compare 2023 and 2024 on revenue drivers and risks'. The agent searches each year's reports separately, notes what they say, and writes a side-by-side summary of what changed, with citations, for a person to check.
Starts: when you run it
Ask a question that people disagree about and get a side-by-side comparison of what several web sources say, where they agree and differ, and a cited synthesis.
Starts: when you run it
Upload exit interview notes or an employee survey export (CSV, PDF, Word or text) and get the recurring themes for each team and across the company, with how often each came up and a paraphrased example. The instructions keep the summary at the aggregate level: no names, no quotes that could identify a person, and teams with fewer than five responses are folded into one group. It is a summary for HR to read, not an assessment of any individual or manager.
Starts: when you run it
Describe your startup and get a first-draft go-to-market plan. A market analyst, a positioning strategist and a copywriter each work on the web search results at the same time, then an editor combines their work into one plan for you to check and adapt.
Starts: when you run it
Answers questions from your documents. Rewrites the question for search, retrieves the best passages from a knowledge base, and replies with linked citations.
Starts: when you run it
Every day at 12:00 UTC, searches the news for each stock or company on your watchlist, summarizes what was reported with links and a label for the tone of the coverage, then combines the summaries into one readable brief that is posted to your webhook. It covers news only, not prices, and it is a reading summary, not investment advice.
Starts: on a schedule
Paste a batch of customer reviews for a product and see what shoppers love, what they complain about, the recurring themes with how often each comes up, short quotes as evidence, and fixes worth considering.
Starts: when you run it
Email a link to an article you don't have time to read and get a reply with the key points, how long the full piece is, and whether it's worth reading in full.
Starts: when an email arrives
Upload a PDF report or article and get a spoken, podcast-style narration of it as an audio file, along with the script that was read.
Starts: when you run it
Upload a research paper as a PDF and get a structured summary: the question it asks, the method, the key findings with numbers, the limitations the authors note, and open questions.
Starts: when you run it
Ask a broad question. A planner breaks it into a few sub-questions, each one is researched on the web at the same time, an analyst weighs what was found and where sources disagree, and a writer turns it into a report with linked sources.
Starts: when you run it
Checks an RSS feed every day at 07:00 UTC, keeps the items about your topics, summarizes each one, and emails you a digest. Sends nothing on days with no matches.
Starts: on a schedule
Give it a company name or website and get a one-page brief before your call: what the company does, recent news, likely priorities, and questions to ask, with a link for every fact.
Starts: when you run it
Every Monday at 08:00 UTC, retrieves the past week's content from a knowledge base and emails you a structured report.
Starts: on a schedule
Every day at 11:00 UTC, searches the web for your topics, writes a short radio-style briefing, and emails it as text and audio.
Starts: on a schedule
Paste a URL or some text and get a three-bullet summary. Branches on the input: a URL is fetched first, anything else is summarized directly.
Starts: when you run it
Paste open-ended survey answers or upload them as a CSV and get the main themes, roughly how many responses mention each, the overall sentiment, and representative quotes.
Starts: when you run it
Enter a vendor's name and get a summary of what public web sources say about it: who they are, recent news, security incidents or breaches, lawsuits or regulatory actions, and financial health signals, each with a link. A starting point for your due diligence, not a risk rating.
Starts: when you run it
Searches the web for your question, reads the most relevant pages, and writes a sourced answer with links.
Starts: when you run it
Paste a product announcement, press release or blog post and get a post written for each network's style and length: LinkedIn, X, Instagram, Facebook and Threads.
Starts: when you run it
Paste an OpenAPI snippet, or an example request and response, and get readable reference documentation: what the endpoint does, its parameters and fields in a table, error responses, and a curl example.
Starts: when you run it
Upload a recording, such as a meeting, voice memo or podcast, and get a short summary back as text and as spoken audio.
Starts: when you run it
Paste a blog post and get a header illustration in one fixed house style. Each image is checked against the style and the post; one that falls short is regenerated, up to three images in total, and you see the first that passes. If none passes, the run ends without a result; the run shows the last image and its score, earlier images are not kept, and every image is charged.
Starts: when you run it
Give it a topic or working title and it searches the web, reads the best sources, and writes an outline plus a full first draft of a blog post in your brand voice, with the sources it used linked for fact-checking.
Starts: when you run it
Upload a few images that show your brand's look (past ads, product shots, your style guide's sample imagery) with a campaign brief, and get four campaign image concepts generated to match that look.
Starts: when you run it
Send each new comment or review from your site and the agent publishes it, publishes it with personal data redacted, labels it rejected and posts nothing, or holds it for a person to approve. Published comments are posted to your site's webhook, and held comments a person approves go to a separate webhook.
Starts: when you run it
Writes a draft from your brief, critiques it against your criteria, rewrites it to fix every point, then scores the rewrite. If the score is below the bar it starts again with a fresh draft, up to three attempts in total, and shows the first rewrite that passes. If none of the three passes, the run ends without a result; the run shows the last attempt and its score, earlier attempts are not kept, and every attempt is charged.
Starts: when you run it
Every Friday at 15:00 UTC, reads up to 15 files changed this week in a cloud drive folder, summarizes each one, and emails you a single digest.
Starts: on a schedule
Paste a customer question and its answer and get a short video clip that shows the idea visually, for your help center or social channels. The clip has no on-screen text, so pair it with the written answer.
Starts: when you run it
Give a story idea, the child's name or age, and anything the story should include. You get a six-page picture-book story and one illustration per page, with the characters described the same way on every page so they look alike.
Starts: when you run it
Upload images and get accessible alt text and a longer description for each one.
Starts: when you run it
Scores an image and its accompanying text: how likely the image is a stock photo or AI generated, how well image and text match, and whether it suits all audiences. Low-confidence ratings go to a person for review.
Starts: when you run it
Upload one or more reference images with an instruction, and get three different takes on it, each generated from your references.
Starts: when you run it
Email an instruction with reference images and get three different variations back by reply. Reply with feedback to get three more; attach the images to keep working from, since each round generates from the images in the latest email.
Starts: when an email arrives
Type a seed topic. The agent searches the web, reads the top-ranking pages, groups related keyword ideas by search intent, and writes a content brief for a writer: angle, headings, questions to answer and the gaps the current results leave. Keyword ideas come from what the pages cover; it has no search volume or difficulty data.
Starts: when you run it
Translates your script into Spanish, French and German, or the languages you set, and records a voiceover for each one.
Starts: when you run it
Paste the links you want to share this week, with optional notes, and get a ready-to-edit email newsletter: a subject line, an intro, and a short summary of each article with its link.
Starts: when you run it
Upload a photo and say who the card is for, the occasion and a few personal details. You get a short card message and a card illustration based on your photo.
Starts: when you run it
Upload a photo, optionally with a question about it, and get a spoken description written for listening: what the scene is, who and what is in it, and any readable text. You get the description as text and as an audio file.
Starts: when you run it
Paste your episode notes and sponsor details and get three scripts, an episode intro, an outro and a sponsor ad read, each recorded as an audio file you can drop into your edit. The ad read only uses claims from the sponsor details you give.
Starts: when you run it
Paste an episode transcript and get show notes: a title idea, a short description, key takeaways, resources mentioned, and a chapter list with timestamps when the transcript has them.
Starts: when you run it
Upload product photos with your specs or supplier notes and get a ready-to-edit listing: a product title, five benefit-led bullet points, a description paragraph and search keywords, written to stick to your specs.
Starts: when you run it
Describe what you are launching and get a matching set: a social announcement post, a launch email with subject line, a header image and a short spoken audio teaser, all written from the same brief.
Starts: when you run it
Describe a product or event and get a short promotional video clip, with the shot written for you first.
Starts: when you run it
Paste a list of merged pull requests, commit messages or ticket titles and get release notes grouped into new features, improvements and fixes, written for your users, with internal-only changes left out.
Starts: when you run it
Upload a PDF report or article and get a spoken, podcast-style narration of it as an audio file, along with the script that was read.
Starts: when you run it
Paste a blog post or its URL and get three ready-to-edit versions: a social media thread, a newsletter blurb, and a two-sentence summary for previews and link posts.
Starts: when you run it
Upload a screen recording of someone doing a task in an app and get a written how-to guide: what it is for, what you need first, and numbered steps naming the buttons and fields used, with the timestamp where each step happens.
Starts: when you run it
Paste a list of page URLs and get, for each page, a search-friendly title aimed at about 60 characters and a meta description aimed at about 155 characters written from what the page actually says, ready to copy into your CMS.
Starts: when you run it
Upload a hand-drawn sketch, whiteboard drawing or wireframe and say what it is; get back a polished render that keeps your layout, plus the brief used to make it.
Starts: when you run it
Drafts a social media post and an illustration from your brief, waits for a person to approve the post text, then sends the approved post and its image to your publishing webhook. The approval request shows the text only.
Starts: when you run it
Every day at 11:00 UTC, searches the web for your topics, writes a short radio-style briefing, and emails it as text and audio.
Starts: on a schedule
Paste a URL or some text and get a three-bullet summary. Branches on the input: a URL is fetched first, anything else is summarized directly.
Starts: when you run it
Every Monday at 08:00 UTC, searches the web for recent news and discussion in your niche and sends up to ten content ideas to your webhook, each with a hook, a suggested format and the link that inspired it. Set your niche in the trigger's input.
Starts: on a schedule
Watches an uploaded video and its caption and decides allow, review or block against your rules, listing the moments that raised concern. Borderline videos go to a person to approve or reject.
Starts: when you run it
Upload a voice message and name the language you want; get back the original transcript, the translation, and the translation read aloud as an audio file. With no language named, it translates into Spanish.
Starts: when you run it
Upload a bank or card statement PDF and get every transaction as a row with date, description, amount and a spending category you define, as a CSV table you can paste into a spreadsheet. The model reads the rows and then a second pass adds them up again and compares the result with the statement's opening and closing balances, listing any mismatch. The arithmetic is done by the model and checked once, so treat the table as a draft to reconcile, not as your books.
Starts: when you run it
Upload a screenshot of something broken, add a line about what you were doing, and get a structured bug report as JSON: a title, steps to reproduce, expected and actual behaviour, error text read from the screen, the environment visible in the screenshot, and a suggested severity.
Starts: when you run it
Paste a supplier's product line, spreadsheet row or spec sheet text and get standard catalog fields back as validated JSON: brand, product type, color, size, material, dimensions in centimetres, weight in kilograms and barcode. The JSON is posted to a webhook you choose (for example an automation tool that updates your catalog).
Starts: when you run it
Paste a list of company domains or upload a CSV. The agent reads each company's home page and records its industry, what it sells, size signals and how well it fits your ideal customer profile, with the reason. Each record is posted as JSON to your webhook, and you get a summary table at the end. A company whose website cannot be processed is listed as not processed instead of stopping the run.
Starts: when you run it
Upload or paste a small CSV with inconsistent dates, stray spaces, mixed capitalization or duplicate rows, and get a cleaned CSV, a list of the changes made, and a short description of each column.
Starts: when you run it
Paste your study notes or a textbook section and get a deck of question-and-answer flashcards as JSON, ready to import into a flashcard app or spreadsheet.
Starts: when you run it
Give suppliers an address to email invoices to. Each PDF invoice is read for supplier, invoice number, dates, PO number, line items and totals, and the fields are posted as JSON to a webhook you choose (for example an automation tool that updates your accounts payable system), for someone to check before payment.
Starts: when an email arrives
Turns unstructured text such as emails, notes or invoices into a JSON object with the fields you define, validated against a schema.
Starts: when you run it
Paste a meeting transcript and get every follow-up meeting, deadline and reminder it agreed on as calendar events in JSON: title, date, start and end time, attendees and notes, ready to import or send to your own tooling.
Starts: when you run it
Upload a PDF and get its text back with names, contact details, ID numbers and other personal data replaced by labels like [NAME], plus a list of what was removed and where, for a person to check before sharing.
Starts: when you run it
Forward supplier order confirmations, purchase orders or delivery notices by email, in the body or as a PDF, and the order number, supplier, dates, line items and total are sent as JSON to a webhook you choose (for example an automation tool that updates your inventory or a spreadsheet).
Starts: when an email arrives
Upload a photo or PDF of a receipt or invoice and send its merchant, date, totals and category as JSON to a webhook you choose (for example an automation tool that updates a spreadsheet or your expense tool).
Starts: when you run it
Paste a call transcript or your rough notes and get the deal fields (stage, budget, decision makers, pain points, next step and date) as JSON, shown to you and posted to a webhook you choose (for example an automation tool that updates your CRM).
Starts: when you run it
Upload a small CSV and ask about it in plain words, then keep asking follow-up questions in the same conversation. The model copies the uploaded table into the conversation, so check its first answer against your file; follow-up questions can refer to the table until it falls outside the roughly 50 most recent exchanges. It suits small files: the model reads the rows and does the arithmetic itself, so check totals that matter.
Starts: when you run it
Paste your table definitions and ask a question in plain words. The agent writes a SQL query, checks that every table and column it uses exists in your schema, fixes what does not, and scores the result; if the score is below the bar it starts again, up to three attempts in total. Nothing is run against a database: you get the query to review and run yourself. If no attempt passes, the run ends without a result.
Starts: when you run it
Paste open-ended survey answers or upload them as a CSV and get the main themes, roughly how many responses mention each, the overall sentiment, and representative quotes.
Starts: when you run it
Email a voice memo and get a reply with a written summary, the action items, and a short spoken recap.
Starts: when an email arrives
Receives JSON or text, extracts the fields you need, checks them, and forwards valid records to your webhook. Invalid records are reported instead of sent.
Starts: when you run it
Send it an alert payload from your monitoring tool. It rates the severity, writes a one-line summary, and forwards critical alerts to your paging webhook and warnings to a second webhook you choose. Informational alerts are only shown in the run result.
Starts: when you run it
Every day at 07:00 UTC, searches the web for your brand or product name and posts the results that mention it to your webhook, each with a one-line summary and whether the mention reads as positive, neutral or negative. Posts nothing when no result mentions it. The same page can come up on more than one day.
Starts: on a schedule
Every day at 09:00 UTC, reads a product or pricing page and posts to your webhook on each day the price of the item you name is below your target. It does not remember earlier prices, so it posts every day the price stays below target. A price it cannot find on the page also posts.
Starts: on a schedule
Every Monday at 08:00 UTC, reads the competitor pages you list (pricing, home or product pages), summarizes each one's plans, prices and headline claims, and posts a side-by-side snapshot with talking points to your webhook.
Starts: on a schedule
Emails you a short summary whenever content is added to or updated in a knowledge base.
Starts: when documents are added or change
Paste an alert or error and your repository. Using your GitHub MCP connection, the agent looks at recently merged pull requests and commits, reads what they changed, and ranks the ones most likely to be related, with its reasoning, as leads for you to check. It needs the GitHub MCP tools list_pull_requests, get_pull_request, get_pull_request_files and list_commits. Tool names are not checked against your server: the agent still runs without any it cannot find, and it says so in its answer when it could not read pull requests.
Starts: when you run it
Scores each customer message for frustration and for signs the customer may cancel or leave, such as competitor mentions or cancellation threats. Messages above your threshold are posted to a webhook you choose, with the reasons and a suggested next step.
Starts: when you run it
Every morning at 08:00 UTC, looks through a repository's open GitHub issues using your GitHub MCP connection and emails you what needs attention.
Starts: on a schedule
Every Monday at 07:00 UTC, reads a regulator's or standards body's news page, summarizes the updates posted in the past week and what they may mean for your team, and posts the summary to a webhook you choose. Posts nothing in a week with no updates.
Starts: on a schedule
Checks an RSS feed every day at 07:00 UTC, keeps the items about your topics, summarizes each one, and emails you a digest. Sends nothing on days with no matches.
Starts: on a schedule
Every Monday at 08:00 UTC, retrieves the past week's content from a knowledge base and emails you a structured report.
Starts: on a schedule
Every Monday at 06:00 UTC, reads a security advisory feed, looks up each advisory from the past week in a knowledge base holding your software inventory, and posts only the ones that appear to match something you run to a webhook you choose. Posts nothing in a week with no matches. Matching is done by AI against your inventory documents, so treat it as a first pass, not a complete scan.
Starts: on a schedule
Send a security alert as JSON from your detection tool through the API. The agent suggests the MITRE ATT&CK tactic and technique the alert most likely represents, lists first triage steps for the analyst, and posts the enriched alert to a webhook you choose. The mapping comes from the model's own knowledge of ATT&CK, with no ATT&CK database attached, so an analyst should confirm it.
Starts: when you run it
Checks a vendor's public status page every hour and, whenever it reports an outage, degraded performance or maintenance in progress, posts a short summary to a webhook you choose. It posts again each hour the problem is still listed.
Starts: on a schedule
Paste a customer message and the reply your chatbot or AI agent wrote, and get it scored from 1 to 5 for accuracy, tone and policy compliance against the rules in the prompt, with the specific problems found. Replies that score low are sent to a person to mark as acceptable or needing a fix.
Starts: when you run it
Checks an image against your brand guidelines. Images the check passes are marked approved without review; anything it flags or is unsure about goes to a person.
Starts: when you run it
Send each new comment or review from your site and the agent publishes it, publishes it with personal data redacted, labels it rejected and posts nothing, or holds it for a person to approve. Published comments are posted to your site's webhook, and held comments a person approves go to a separate webhook.
Starts: when you run it
Scores an image and its accompanying text: how likely the image is a stock photo or AI generated, how well image and text match, and whether it suits all audiences. Low-confidence ratings go to a person for review.
Starts: when you run it
Reads a seller's listing text and photos and checks them for prohibited items, scam signals and misleading claims, then approves, rejects or sends the listing to a person to review.
Starts: when you run it
Checks an uploaded profile photo: does it show one real person's face, is it free of text, logos and contact details, and is it suitable for your site. Accepts or rejects it with a reason, and asks a person when it is unsure.
Starts: when you run it
Tells you whether to block, redact or allow a message: returns a blocked notice, the redacted message, or the message unchanged.
Starts: when you run it
Watches an uploaded video and its caption and decides allow, review or block against your rules, listing the moments that raised concern. Borderline videos go to a person to approve or reject.
Starts: when you run it
Listens to a voice message and returns a transcript, the exact phrases that break your rules, and a verdict: allow or block, with the reason. Accepts MP3, WAV, AAC, OGG and FLAC recordings.
Starts: when you run it
Send an employee's access request in their own words. The agent checks it against the access policy you write into the prompt (least privilege, a stated business reason and a time limit) and passes that assessment to an owner or admin as advice. Every request needs a person's approval: only requests they approve are sent to your provisioning webhook. The check reads only the request text, so the requester named in the request is not verified, and nothing looks up the person's role or current access.
Starts: when you run it
Staff forward a suspicious email, with screenshots if they have them, and get a reply rating it phishing, suspicious or looks safe, with the reasons: sender and header oddities, where the links really go, and pressure tricks. The agent opens only the main site of each linked domain (for example https://example.com), never the link itself, to see what is there. This is an AI review, not a reputation or blocklist lookup, and "looks safe" is not a guarantee.
Starts: when an email arrives
Every Monday at 06:00 UTC, reads a security advisory feed, looks up each advisory from the past week in a knowledge base holding your software inventory, and posts only the ones that appear to match something you run to a webhook you choose. Posts nothing in a week with no matches. Matching is done by AI against your inventory documents, so treat it as a first pass, not a complete scan.
Starts: on a schedule
Send a security alert as JSON from your detection tool through the API. The agent suggests the MITRE ATT&CK tactic and technique the alert most likely represents, lists first triage steps for the analyst, and posts the enriched alert to a webhook you choose. The mapping comes from the model's own knowledge of ATT&CK, with no ATT&CK database attached, so an analyst should confirm it.
Starts: when you run it
Upload a customer's security questionnaire as a PDF or Word file. Each question is answered from your security policies and documents in a knowledge base, with the source document linked, or marked "Needs input" when your documents do not cover it. The result is a draft for a person to check before it goes to the customer.
Starts: when you run it
Paste an application or upload a CV and up to six employers it names (the most recent first) are searched on the web. You get a list of things worth checking (employers with no web presence, dates that overlap or don't add up, job titles that don't match the company) and a low, medium or high level of concern. It never rejects anyone: a recruiter reads the notes and decides what to verify.
Starts: when you run it
Paste a web address and get a risk level (high, medium or low) with the reasons. The agent searches the web for scam reports and reviews of the domain and tries to read the site's main page (https:// plus the domain), then weighs what it found. When the site cannot be loaded, the rating rests on the search results alone. This is an AI review of public information, not a reputation database lookup, so a low rating is not a guarantee.
Starts: when you run it
Give customers an address for billing questions. Each email is sorted into remittance advice, invoice dispute, invoice copy request, tax form request or other, and the invoice number and amount are pulled out. Each of the four billing types is posted to its own webhook; an email sorted as other is not posted anywhere. A suggested reply is drafted for a person to check and send; nothing is sent to the customer.
Starts: when an email arrives
Upload a recording of a sales or support call and get CRM-ready fields as JSON (contact, company, call type, summary, needs, objections, next step, follow-up date) plus the customer's sentiment, sent to a webhook you choose. Accepts MP3, WAV, AAC, OGG and FLAC recordings.
Starts: when you run it
Forward customer replies on support tickets to the agent. Replies that only say thanks or confirm the issue is fixed are sent to your close-ticket webhook; anything that asks for more, reports a new problem or is unclear goes to your support queue webhook instead.
Starts: when an email arrives
Send each new comment or review from your site and the agent publishes it, publishes it with personal data redacted, labels it rejected and posts nothing, or holds it for a person to approve. Published comments are posted to your site's webhook, and held comments a person approves go to a separate webhook.
Starts: when you run it
Paste or upload a purchase order or booking request and its details (customer, items, quantities, prices, dates, delivery address) are pulled out and shown to a person field by field. They are posted to your webhook only if that person approves. The reviewer cannot edit fields here: rejecting (or not answering within a day) sends nothing, so fix the source document and submit it again.
Starts: when you run it
Every Friday at 15:00 UTC, reads up to 15 files changed this week in a cloud drive folder, summarizes each one, and emails you a single digest.
Starts: on a schedule
Every morning at 08:00 UTC, looks through a repository's open GitHub issues using your GitHub MCP connection and emails you what needs attention.
Starts: on a schedule
Email an instruction with reference images and get three different variations back by reply. Reply with feedback to get three more; attach the images to keep working from, since each round generates from the images in the latest email.
Starts: when an email arrives
Reads a contact-form or demo-request submission, pulls out the contact details, scores how well the lead fits your ideal customer, and sends leads scoring 70 or more to your sales webhook. Lower scores are shown with the reason instead.
Starts: when you run it
Claimants email their claim and supporting files to an address you give them. The email and attachments are checked against your list of required items. If something is missing, the claimant gets a reply listing what is missing. Each email is checked on its own, so the reply asks the claimant to send the whole claim again in one email, with every document, including the missing ones. If everything is there, the claim details are posted to your claims system's webhook for a handler. It checks that items are present, not whether a claim is valid or covered.
Starts: when an email arrives
Give suppliers an address to email invoices to. Each PDF invoice is read for supplier, invoice number, dates, PO number, line items and totals, and the fields are posted as JSON to a webhook you choose (for example an automation tool that updates your accounts payable system), for someone to check before payment.
Starts: when an email arrives
Employees email IT with a problem, optionally with screenshots. The agent categorizes the request, rates its urgency, sends the triaged ticket as JSON to your helpdesk webhook, and replies to the employee with self-help steps to try while they wait.
Starts: when an email arrives
Forward supplier order confirmations, purchase orders or delivery notices by email, in the body or as a PDF, and the order number, supplier, dates, line items and total are sent as JSON to a webhook you choose (for example an automation tool that updates your inventory or a spreadsheet).
Starts: when an email arrives
Forward your support inbox to the agent. It labels each email with a category, urgency and customer sentiment, posts the labels to your helpdesk webhook, and emails you straight away when an email is urgent.
Starts: when an email arrives
Send each new ticket from your helpdesk or contact form through the API. The agent decides whether it belongs to billing, technical support or account management, sets a priority, and posts it to that team's webhook; anything else goes to a general queue.
Starts: when you run it
Employees email their vacation or leave requests to an address you give them. The dates and type of leave are pulled out and a manager or admin is asked to approve or reject. An approved request is posted to your HR system's webhook and the employee gets a reply saying it was approved; a rejected one gets a reply saying so. If nobody answers within three days, nothing is sent.
Starts: when an email arrives
Email a voice memo and get a reply with a written summary, the action items, and a short spoken recap.
Starts: when an email arrives
Receives JSON or text, extracts the fields you need, checks them, and forwards valid records to your webhook. Invalid records are reported instead of sent.
Starts: when you run it