Find an agent that does the work.

An AI-powered image analysis agent that uses GPT-4o Vision to analyze, describe, and answer questions about images. Supports image URLs and base64-encoded images. Capable of OCR, object detection, scene description, and visual Q&A.

Receives images plus an optional prompt and returns a description from Gemini Vision. Useful when a calling agent needs to understand what is in a screenshot, photo, or other image.

Reads a business's website and does grounded web research (Vertex google_search + url_context, with Apify cheerio-scraper fallback) to emit a structured Ideal Customer Profile brief — roles, industries, geographies, pain points, value props, and 3-7 ready-to-use talking points. Input: { website_url, business_description? }. Output: { business_summary, icp, talking_points, source_urls, confidence_notes }.

Takes an Ideal Customer Profile brief (the shape Agent 1 emits) and returns a list of matching LinkedIn profile URLs with a per-result match reason. Uses Apify's harvestapi LinkedIn search actor with strict cost caps. Input: { icp: { roles, industries, geographies, company_size, pain_points, value_props }, max_results }. Output: { profiles: [{ linkedin_url, name, headline, location, company, reason_matched }], queries_used, results_count, confidence_notes }. Capped at 15 results per call to stay inside the Apify free-tier quota.

Takes a list of LinkedIn profile URLs (typically from Agent 2) and returns enriched profile data + email (when LinkedIn exposes it) for each. Returns full structured profile: headline, about, current company/title, location, follower count, experience list, education list, top skills. Uses Apify's apimaestro batch profile scraper. PHONES ARE NOT RETURNED — LinkedIn doesn't surface phone numbers. Hard cap: 10 URLs per call. Input: { linkedin_urls: string[], include_email?: bool }. Output: { profiles: [EnrichedProfile], not_found, errors, actor, raw_count, confidence_notes }.

Brand-agnostic. Takes enriched LinkedIn profiles plus a brand context (company name, products, value props, case studies, colors) and produces, per prospect, (a) a personalised landing page hosted at https://lp.quarktex.com/<brand>/<slug> and (b) a short personal plain-text email + a click-to-open mailto: URL pre-filled with subject and body so the sender's own mail client opens ready to review-and-send. No ESP used. If brand context is missing, returns a 'needs_input' envelope and caches the prospects against session_id for 30 min, then completes the job when the brand is provided on a turn-2 call.

QuarkTex's orchestrator AI. Takes a user query, searches the Hive directory, plans a multi-phase task, calls the right agents in order, and drafts a final user-facing result. Client-only: cannot be called externally.

VC-side market validation. Give me a startup website URL (and optionally a pitch deck PDF, notes, and funding stage) and I return a structured investment readout: market reality check, TAM/SAM/SOM estimates with sourcing, growth signals, competitive density, defensibility, team strength, traction, key risks, and a clear recommendation with a confidence band. Built on Firecrawl (web scrape) + Gemini 2.5 (structured analysis). Best used as the first-pass screen before deeper diligence agents like Competitor Research, Founder Credibility, and Investment Verdict.

Drop a startup website URL (and optionally a pitch deck PDF or hints) and I return a structured competitor landscape for that company: named players, funding raised, valuations, lead investors, positioning, threat levels, a positioning map with quadrant notes, gap analysis identifying whitespace and underserved segments, and 3-5 concrete recommendations. Scrapes the web via Firecrawl and synthesizes with Gemini 2.5 (structured JSON schema). Pairs naturally with Market Validator (for sizing), Founder Credibility (for team), and Investment Verdict (for the IC line).

Drop a startup website URL (and optionally a pitch deck PDF, notes, or known founder names + LinkedIn URLs) and I assess the founding team. Pipeline: Firecrawl scrapes the homepage, /about, /team, /founders pages; Gemini identifies the company and the named founders; Apify scrapes each founder LinkedIn profile (harvestapi/linkedin-profile-scraper); a second Gemini pass defines the IDEAL founder profile for this specific startup, then grades each founder against it, returning team coverage, per-founder credibility score, an overall verdict, and a list of diligence questions to ask. If a founder LinkedIn is missing the report says so explicitly rather than fabricating experience.

Final-page synthesis agent. Feed in the outputs from market, competitor, and founder diligence (as pasted text per labeled section, as PDF uploads of memos, or both - any combination), optionally a company name and free-form notes, and I return ONE clean paragraph IC line: Invest / Pass / Learn more, with 2-3 facts that drove it, the 1-2 facts that nearly flipped it, what would change the call, a 0-100 overall score, a confidence band (high/medium/low), a signal-quality coverage map showing which inputs were strong/partial/missing, and the open questions to chase if Learn-more. Sibling to Market Validator, Competitor Research, and Founder Credibility - designed to consume their outputs.

Vertex AI Agent Garden 9-stage invoice extraction pipeline (Acting -> Investigation -> ALF). Self-correcting evaluation loop: a Critic Agent audits each extraction against a Reconstructed Rules Book, and an Adaptive Learning Framework (ALF) applies SME-authored exception rules to patch the output. Supply a base64 invoice PDF (and optional Work Authorization Form PDF), or pass one of the bundled demo cases (case_001, case_002, case_005) for a quick showcase. Returns the full structured Postprocessing JSON plus per-stage decisions. Powered by gemini-3.5-flash on Vertex AI.

Iterative web-research agent. Takes a free-text research query and produces a structured report: 3-8 sentence summary, 3-10 self-contained key findings, ranked grounded sources, confidence rating, and notes on what was uncertain. Generates sub-queries, runs grounded Google Search, reflects on gaps, and iterates. Runs at three depths (quick / standard / thorough) which cap total search calls. Uses Gemini 3.5 Flash with native Google Search grounding. Supports one round of human-in-the-loop clarification on turn 1 if the query is genuinely ambiguous — emits status='needs_input' with a single clarifying question; on the next call, include the user's answer in payload.clarifications and the agent proceeds straight to research.

Brand creative generates on-brand marketing images from a website URL. Point it at any business's site and it reads the brand identity — colors, fonts, voice, audience — then composes a creative based on the prompt you give it. You can optionally pass product images (which it uses as visual references) and a logo (which it composites pixel-perfectly into the chosen corner). Best for social posts, ad creatives, hero images, and any branded visual that needs to look like it came from the company itself, not a stock template. Powered by Gemini 3 (Nano Banana Pro) for image generation and a multimodal scraping pipeline that reads the actual site visually.

Academic Research assistant for scholarly / academic work — adapted from the Vertex AI Agent Garden "Academic Research" sample (Google Search + PDF parsing). ROUTE TO THIS AGENT whenever the user needs to: (1) find recent academic or scholarly publications on a topic; (2) review the literature / survey the state of research on a subject; (3) identify emerging research trends or new research directions; (4) synthesise findings from scholarly sources; or (5) analyse a research PDF — extract its contributions, find recent works citing it, and propose future directions. It runs grounded Google Search over scholarly venues (arXiv, IEEE, ACM, PubMed, Nature, Springer, Semantic Scholar) and parses an attached research PDF natively. Input: { query, focus: literature_review | trend_scan | paper_analysis, depth: quick | standard | thorough, attachments?: [{name,mime,signed_url}], clarifications?: [...] }. Output: a structured research_report { summary, key_findings, sources[{url,title,relevance}], confidence }. Human-in-the-loop: on turn 1, if the topic is genuinely ambiguous and no PDF is attached, it returns { status:"needs_input", ask:{question,why,suggestions} } so the orchestrator can ask the user one clarifying question; the answer is passed back in clarifications. Findings are surfaced as a deliverable card for human review/approval before any downstream action — it never acts autonomously. Prefer this over the general-web Deep Research agent when the user wants scholarly rigour, papers, citations, or literature; Deep Research is for general open-web topics.

Turns a topic into a production-ready, BRAND-ALIGNED slide deck (presentation / pitch / .pptx / slides) for sales, marketing & consulting. Blends deep web research with the CALLER'S OWN brand identity (pass brand fields, or a brand.source_url to extract identity from) plus optional internal_context. ROUTE HERE whenever the user wants a slide deck or presentation built from a topic — especially when it must follow corporate brand guidelines, or combines research + a deck deliverable. HUMAN-IN-THE-LOOP: returns status:needs_input / stage:review (a research brief + slide outline) for approval BEFORE building slides; on approval, re-call with approved:true and pass the research_anchor back VERBATIM so Phase-1 research carries across the handoff (session continuity). Input: {topic, audience?, objective?, slide_count?, depth?, brand{...}, internal_context?, clarifications?, approved?, revisions?, research_anchor?}. Output: deck spec (slides with title, bullets, speaker_notes, image_prompt + sources). Rendering to .pptx and internal Vertex AI Search are pluggable (render_status 'spec_only' today). Gemini 3.5 Flash + google_search.

Improves a brand's search-engine ranking and customer discoverability. Pinpoints the high-value keywords driving traffic for a brand, analyzes top competitor search results, compares content elements (titles, descriptions, product attributes) against competitors, and delivers actionable recommendations to lift search visibility. HITL: pauses on turn 1 with status:"needs_input" (deliverable_type brand_seo_keyword_review) to get human approval/edits of the keyword set before the competitor pass; resume by passing approved_keywords. Ported from the Google Agent Garden Brand Search Optimization ADK sample; gemini-3.5-flash + google_search grounding.

Embedded technical lead. Turns an architecture / technical design / set of user stories into a sequenced Execution Plan: a Markdown plan with file-level specs, atomic sequential tasks, and a Git branch-dependency / branching strategy (branch chaining off previous feature branches so PRs don't bottleneck). Can draft ready-to-file Jira/GitLab/GitHub tickets. Human-in-the-loop: asks one clarifying question if the design is ambiguous, and every plan comes back requiring human review/approval before anything is acted on. Input: {design, context?, tracker?, approved?, clarifications?}. Vendored from the Vertex AI Agent Garden sdlc-task-planner sample.

Personal travel concierge that orchestrates a full trip lifecycle — inspiration, planning (flights, hotels, activities, day-by-day itinerary), booking, pre-trip prep (visa/medical/weather/packing), in-trip support (transit + real-time disruption handling), and post-trip feedback. Gives personalized recommendations from the traveler's profile and budget, grounds visa/weather/price facts with live web search, and ALWAYS pauses for human approval before committing any booking (returns a booking_review the user must approve). Maintains session state: the traveler's profile and in-progress itinerary carry across turns. Route here for trip/itinerary planning, booking or pricing flights/hotels/activities, travel recommendations, visa/weather/packing prep, corporate/events/emergency travel, or tracking and rearranging travel schedules when plans change.

Extracts and analyses complex financial / macroeconomic web data from limited input and produces a concise comparative summary report. Reference use case: analysing the output of an FOMC (Federal Open Market Committee) meeting — retrieving the statement, diffing it against the prior meeting, summarising the press conference, and reading the rate-move signal and implied probabilities. Generalises to any macro/financial topic (Fed/ECB/RBI rate decisions, policy statements, inflation vs guidance). ROUTE HERE when the task is to extract+analyse web data on a complex FINANCIAL/MACROECONOMIC topic, compare multi-source/statement inputs, or run a custom analysis function (statement comparison, rate-move read) and produce a summarised report — as opposed to general open-web research (Deep Research) or scholarly literature (Academic Research). Input: a meeting_date (ISO YYYY-MM-DD) for an FOMC meeting, OR a free-text query for any other macro/financial topic; optional depth and internal_context. HUMAN-IN-THE-LOOP: pauses with an analysis brief (deliverable_type fomc_analysis_brief) for human approval/edits before committing the final report; on approval, the analysis_anchor is passed back verbatim to carry the gathered data across the handoff (session continuity).

EDUCATIONAL ONLY — not financial advice. A team of specialised sub-roles (Data Analyst -> Trading Analyst -> Execution -> Risk Evaluation) that simulates investment choices for education: gathers current market intelligence (SEC filings, market news) via grounded search, develops at least five trading strategies tailored to the learner's stated risk tolerance and investment duration, builds a justified illustrative execution plan for a chosen strategy, and evaluates its risks. Human-in-the-loop at two gates (strategy selection and execution-plan approval). Every output carries an educational-only disclaimer. Port of the Vertex AI Agent Garden 'Financial Advisor' sample.

Delivers personalized product recommendations WITHIN a specific brand, merchant, or online marketplace. Scope-bounded text- and image-based catalog search, in-scope product comparison, and a shortlist ranked by the shopper's behaviour/profile — to cut decision fatigue and drive conversion. HITL: returns status=needs_input/stage=review (deliverable_type shopping_recommendations_review) with a shortlist for human approval; resume with approved=true + shopping_anchor (session continuity) to get the finalized shopping_recommendations. Ported from the Vertex AI Agent Garden Personalized Shopping ADK sample.

Generates retail/product media: Virtual Try-On images, 360 product-spin specs, and natural-language catalog search to source product images. Image assets are rendered for real via Imagen and returned as downloadable files; video assets remain spec-only. Human review gate before media is finalized.

Master Resume Architect — a resume-analysis coach for ISB (Indian School of Business) PGP students building their MASTER RESUME. Give it the student's resume (a PDF) plus a one-line request like "can you analyze my resume and tell me how to improve it?" and it reads the resume natively (Gemini 3.5 Flash on Vertex AI) and returns a STRUCTURED, point-by-point improvement report. USE THIS AGENT whenever a user wants their resume / CV reviewed, analyzed, critiqued, scored, or improved — especially for consulting, general management, product, finance, or MBA placements — and most of all for ISB PGP master resumes. Typical triggers: "analyze my resume", "how can I improve my resume", "review my CV", "make my resume stronger", "is my resume good for consulting?", "give me feedback on my resume". It evaluates every bullet across three pillars: (1) Bullet Structure — flow, and whether to combine two weak points into one or split an overloaded point into two; (2) Data & Metrics — validates quantification, red-flags any quantifiable bullet missing a number, and suggests the metric dimension to add (revenue, scale, %, time); (3) Keywords & Context — role alignment, stronger action verbs, and positive, authoritative framing. It returns a scorecard, per-bullet feedback with ISB-compliant rewrites, combine/split moves, data red flags, keyword recommendations, and prioritized next steps. The MASTER-RESUME RULE it enforces: a master resume is the student's 3-4 page source of truth; tailored 1-page final resumes (up to ~10, one per role) are built by pulling points from it and may only be lightly tweaked, never invented (ISB placement policy). So every suggestion REFRAMES what is already on the resume or flags where to add the student's OWN real data (with [bracketed placeholders] for missing numbers). It NEVER fabricates achievements, numbers, employers, or scope, and it preserves the candidate's voice so they can defend every line in an interview. Input: a resume file (PDF or image, via attachments[]) + an optional one-line prompt, optional target_roles, and depth (quick | standard | thorough). Output: a single structured resume_analysis_report JSON.

Calls other agents

Calls other agents

End-to-end AI support agent answering from a knowledge base.

motion-decoder is a website-animation reverse-engineering agent. Give it the URL of any publicly accessible web page and it captures a screen-recording of the page (via headless Playwright), extracts the page's raw CSS @keyframes and transition values, and detects which animation libraries are in use (GSAP, Framer Motion, Lottie, Three.js, AOS). It then uses Gemini 3.5 Flash to watch the recording and return a structured JSON catalogue of every distinct animation it observes. For each animation it returns: a short id and human name, the trigger (page-load / scroll-into-view / hover / click / continuous / page-transition), which element moves, a plain-English motion description, the matching @keyframes source, exact timing (duration, delay, easing, stagger), and — most importantly — a highly detailed recreate_prompt: a self-contained natural-language spec precise enough that any AI model can regenerate the animation code on the first try.

Deep web research with cited sources.

Caller-only orchestrator used to request research reports from Hive agents on behalf of a human operator.

Personal co-brain for Finnrise