--- type: "nimble-use-case" name: "deal-sourcing" description: "Nimble's deal sourcing agent populates a ranked list of private companies that fit your investment thesis, with each company's ownership, size, growth signals and traction, from the live web. It searches funding and growth lists, company databases, trade and local business press and the companies' own sites, and gives the evidence for every criterion and the reason for each rank." use_when: "Turning an investment thesis into a ranked list of private companies, with the evidence for every criterion." not_for: "Non-public financials, deal flow from brokers, proprietary databases of private company revenue." agent_id: "wsa_ed77b118ad644c50a541defa46fc5958" price_per_agent_run: { low: 0.025, medium: 0.1, high: 0.5 } updated: "2026-11-11" --- # AI web research agent for deal sourcing (Nimble) Nimble's deal sourcing agent populates a ranked list of private companies that fit your investment thesis, with each company's ownership, size, growth signals and traction, from the live web. It searches funding and growth lists, company databases, trade and local business press and the companies' own sites, and gives the evidence for every criterion and the reason for each rank. - Human page: https://www.nimbleway.com/use-cases/finance/deal-sourcing - Interactive demo: https://www.nimbleway.com/demos/deal-sourcing - Agent context: https://www.nimbleway.com/use-cases/finance/deal-sourcing/deal-sourcing-agent.md - Skill for agents: https://docs.nimbleway.com/skills/deal-sourcing/SKILL.md - Docs: https://docs.nimbleway.com/nimble-sdk/web-search-agents/overview ## 1. Choose this when - You have a thesis and want companies that meet every criterion, not a long list of near misses. - Ownership matters: founder-owned, family-owned, PE-backed (with the firm) or VC-backed. - You are looking for add-ons for a platform and want local and niche companies that databases miss. - You want each company ranked with a reason, and the evidence for each criterion linked. ## 2. Use something else when - You need private revenue or EBITDA. Public pages rarely have them; the agent reports only what a source states. - You want a deep read on one company. Use the due diligence agent. ## 3. Minimal request One run of the ready-made deal sourcing agent (`wsa_ed77b118ad644c50a541defa46fc5958`). ```bash AGENT_ID=wsa_ed77b118ad644c50a541defa46fc5958 # Nimble's deal sourcing agent curl -X POST "https://sdk.nimbleway.com/v2/agents/$AGENT_ID/runs" \ --header "Authorization: Bearer $NIMBLE_API_KEY" \ --header 'Content-Type: application/json' \ --data '{"input":"Find bootstrapped European vertical SaaS companies for logistics or freight with 50-200 employees","effort":"high"}' # poll until is_active is false, then read the result curl "https://sdk.nimbleway.com/v2/agents/$AGENT_ID/runs/$RUN_ID" --header "Authorization: Bearer $NIMBLE_API_KEY" curl "https://sdk.nimbleway.com/v2/agents/$AGENT_ID/runs/$RUN_ID/result" --header "Authorization: Bearer $NIMBLE_API_KEY" ``` The result is in `output.content`, a list of `{ companies, search_notes }`. Each company has `rank`, `name`, `domain`, `hq`, `employees`, `ownership`, `founded_year`, `last_round`, `headcount_growth`, `traction`, `rank_reason` and `criteria_evidence` (a list of `{ criterion, evidence, source_url }`). ## 4. Use it from an agent Paste this into Claude, ChatGPT, Codex, Cursor or Grok: ```text Be my deal sourcing assistant using Nimble. Follow https://docs.nimbleway.com/skills/deal-sourcing/SKILL.md. ``` The full skill is in section 7 below. - Frameworks: `@nimble-way/mastra`, `langchain-nimble`, `@nimble-way/ai-sdk`, `llama-index-tools-nimble`, `nimble_python`, `@nimble-way/nimble-js`. ## 5. Example runs ### European logistics SaaS (recorded 2026-11-11) Input: Find bootstrapped European vertical SaaS companies for logistics or freight with 50-200 employees Run record: high effort, 6.5 minutes, 12 pages cited. - #1 TIS GmbH (TIS Technische Informationssysteme GmbH) (tis-gmbh.de), Bocholt, Germany, 110 employees (regional employer profile); 80 employees per company fact sheet dated June 2021, Family-owned/founder-owned: founded 1985 by Josef Bielefeld, who is still the owner together with his son; no PE/VC funding mentioned: Meets every hard criterion with published evidence: European (Bocholt) family-owned logistics/freight vertical software vendor, 110 employees within the 50-200 band, self-funded family growth; strongest headcount-growth evidence of the list - #2 Riege Software International GmbH (riege.com), Meerbusch, Germany, 120+ employees (company press kit), Family-owned: "Riege Software remains a family-owned business" — five second-generation Riege siblings jointly lead the company; no PE/VC funding mentioned: Fully matches thesis: Meerbusch-based family-owned freight-forwarding TMS (Scope, SaaS), 120+ employees in band, decades of self-funded international growth; headcount growth unverifiable (only one published count) - #3 AndSoft (andsoft.com), La Massana, Andorra, 70 collaborators (company site 'In numbers'); 54 employees as of May 2026 (Tracxn); LinkedIn band 51-200 employees, Unfunded/self-funded: "AndSoft is an unfunded company... AndSoft has not raised any funding yet" (Tracxn); founder/owner names not published: Independent Andorra-based e-TMS vendor in the employee band with verified no-funding history; founder identity not published, so ownership rests on the unfunded record - #4 Boltrics (boltrics.com), Ede, Netherlands, 68 employees (LinkedIn employee_count), Founder-owned: Jan Bolks, Founder and Managing Director (CB Insights); no funding/PE/VC records found: Dutch 3PL-focused WMS/TMS/FMS SaaS with 68 employees and founder-ownership on record; no funding round ever published, but founder/bootstrapping evidence is thinner than the top three Search notes: Built the candidate pool from the published European TMS/logistics-software vendor landscape (transportmanagement.org market-by-market roundup, Dec 2025), logistics software directories (ezilon, Cargoson's 2026 TMS guide), VerkehrsRundschau's logistics Who-is-Who, and each vendor's own site, LinkedIn record and registry/ownership databases. Verified headcount, ownership and funding per company before inclusion. Qualifying: TIS GmbH (Bocholt, DE — family-owned by Josef Bielefeld, 1985, 110 employees, up from 80 in 2021), Riege Software (Meerbusch, DE — family-owned, 1985, 120+, Scope SaaS TMS), AndSoft (Andorra — unfunded per Tracxn, 1998, 70 collaborators), Boltrics (Ede, NL — founder Jan Bolks, 2007, 68 employees). Excluded after verification (criterion failed): Soloplan (family-owned but 260 employees), TIMOCOM (family-owned but >700 employees), Trans.eu Group (350 FTE), AEB SE (employee-owned but 800+ employees), AKANEA (subsidiary of Groupe Isagri, 270), TESISQUARE (501-1,000 employees), Solvares Group (330+ employees), Mandata (PE-backed by Tenzing), SINARI (Bridgepoint majority stake 2023), ecovium (FSN Capital since 2020), Inelo Group (acquired by Eurowag 2022), Opter (listed on Nasdaq First North since 2021), Vigo Software (part of Klipboard), HaulTech (11-50 employees, below band), Cargoson (17 employees, below band), Webtrans (seed-funded by BRAIN VC, 11-50). Alpega Group and its brands (Teleroute, Wtransnet, Alerce, Navitrans), Transporeon (Trimble), nShift, AddSecure, Pagero (Thomson Reuters), Microlise (listed) and Aptean 3T (PE group) are groups or external-capital-owned, not founder-owned independents. ### US construction SaaS (recorded 2026-11-11) Input: Find US vertical SaaS companies for construction at seed to Series A, with 10-100 employees Run record: high effort, 6.1 minutes, 25 pages cited. - #1 Digs (digs.com), Vancouver, Washington, US, 37 employees as of Aug 2026 (within 10-100), VC-backed and strategic-led: Series A led by Builders FirstSource; previously backed by Fuse, Flying Fish, Oregon Venture Fund, Cascade Seed Fund: Strongest traction in the qualifying set: a five-year strategic partnership with Builders FirstSource reaching 140,000 builder clients, national footprint across all 50 states, largest round ($25.3M Series A) with headcount squarely in the 10-100 band and growing - #2 SubBase (subbase.io), Fort Lauderdale, Florida, US, 43 employees (within 10-100), VC-backed: Series A led by FINTOP with participation from Fika Ventures; founder/CEO Eric Helitzer: Quantified revenue-volume traction ($1B+ materials volume pace) and a dedicated materials-procurement wedge into subcontractor workflows, with headcount (43) comfortably mid-band - #3 BoreDM (boredm.com), Phoenix, Arizona, US, 20 employees (within 10-100), VC/growth-backed: Series A growth investment from Five Elms Capital; founded by Louis Aaron (Founder & CEO): Clear public-sector foothold (18 state DOTs) in an underserved geotech data niche; small headcount (20) leaves a long growth runway inside the 10-100 band - #4 Buildcheck AI (buildcheck.ai), Palo Alto, California, US, 41 employees (within 10-100), VC-backed: Series A led by Telescope Partners with WND Ventures, Uncork Capital, Salt VC and Xfund; founded at Stanford University by Alexander Michalatos, Andrei Molchynsky and Alex Gureiev: Named enterprise customers (DPR, EllisDon, Power) plus fast round cadence ($5.9M seed to $12M Series A in nine months); headcount (41) mid-band - #5 Quotr.ai (quotr.ai), San Francisco / Berkeley, California, US, 26 employees (within 10-100), VC-backed: $4M seed led by Llama Ventures with strategic angels; product of FLOZ Inc.; started at UC Berkeley out of Batch 19 at Berkeley SkyDeck: Hard usage and revenue proxies (80% takeoff-time reduction, $1.2B construction value estimated) at fresh seed; headcount (26) squarely in band - #6 Scaffold (getscaffold.com), Austin, Texas, US, 50 employees (within 10-100), VC/strategic-backed: $15M seed led by Navitas Capital; investors include homebuilders D.R. Horton and Pulte Homes, Builders FirstSource, Windsor America, Grid Capital, Home Technology Ventures, Stage2 Capital, and housing analyst Ivy Zelman; co-founded 2024 by Ben Johnson (CEO) and Robert Eanes (CTO): Deepest strategic syndicate of any seed in the set (top homebuilders and BFS investing) with 200,000-home deployment footprint; headcount (50) mid-band - #7 Structured AI (getstructured.ai), New York City, US, 11 employees (within 10-100, at the lower bound), VC-backed: $4.2M seed led by FCVC with participation from Y Combinator, 20VC, Cherry Ventures, Zero Prime Ventures, Transpose Platform and Sequoia Capital's Scout Fund; cofounder/CEO Raymond Zhao: AEC design QA/QC niche with a named engineering-firm partnership (Syska Hennessy) and ENR coverage; qualifies on headcount at the lower bound (11) - #8 Primepoint (primepoint.ai), San Mateo, California, US, 18 employees (within 10-100), VC-backed: funded by Navitas Capital, Penny Jar Capital, NextView Ventures, GS Futures, Aglaé Ventures and angel Dr. Yann LeCun; co-founded by AI veterans Lubomir Bourdev, PhD and Hamid Palo: $10M seed (largest of the seeds in the set) with a marquee angel (Yann LeCun) and named GC partner Sundt Construction; young team (18) with room to grow in band - #9 Guthrie AI (guthrieai.com), Philadelphia, Pennsylvania, US, 12-person team as of Jul 2026, planning to add six more Philadelphia-based roles (within 10-100), VC-backed: $4M seed led by Chicago Ventures, first external capital after bootstrapping and a friends-and-family round; founded by CEO Ted Baumgardner with cofounder Tuneer De: Narrow but deep wedge (glazing contractors) with 52 named deployments of its Virtual Bid Assistant and a stated $2M-to-$20M revenue goal; headcount (12) just above the lower bound - #10 TruTec.ai (trutec.ai), San Francisco, California, US, 10 employees as of Sep 20, 2026 (at the lower bound of 10-100), VC-backed: $4.6M seed led by Pear VC; also backed by Aglaé Ventures; PearX W26 cohort company; co-founders Ross Stitts (CEO) and Ryan Brown (CTO): Fast-growing (5 to 10 employees in one quarter) with a dozen named paving customers at seed, but headcount sits exactly at the 10-employee lower bound and the company is the youngest in the set (launched Jan 2026) Search notes: Searched 2025-26 construction-tech funding roundups (Growth List 650+ funded construction startups; VCBacked 454 seed construction companies; SeedTable construction-tech rankings; Cemex Ventures Top 50 Contech 2026; YC construction/real-estate directories; bricks-bytes weekly roundups; Nymbl Q3-2025 contech market report), then verified each candidate's stage from its own funding announcement or press coverage, and headcount from company profile data, YC directory, or dated press. Qualification applied strictly: US HQ, vertical software for construction (robotics/materials/services companies excluded), last round Seed or Series A, 10-100 employees. Ten companies qualify. Excluded on headcount (>100): CivilGrid (110), Agave (101), Buildforce (111), Track3D (156); on headcount (<10): Punch (4), Parcodata (3), ProBlu (2), Captus AI (7); on business model: Gritt AI (construction robotics, not SaaS), Trebellar (corporate real estate), Origin (robotics), Rise Modular/Kind Designs/Everframe (manufacturing); on geography: Truss (Victoria, Canada), SubBase originally checked and confirmed US HQ Fort Lauderdale; on unverifiable stage: SmartBarrel (rounds undisclosed, cannot confirm seed-Series A) and DrawScale (seed confirmed but no published employee count); Miter, PermitFlow, Adaptive, GreenLite, Trunk Tools, Kojo, Kahua, Higharc, PlanHub are past stage; Document Crunch was acquired by Trimble in April 2026 and no longer qualifies. Headcount caveat: Guthrie AI's press coverage (Jul 2026) reports a 12-person team while one data profile lists 77; the dated, specific press figure is used. Digs' profile data lists 60 employees while GeekWire (Aug 2026) reports 37 with a plan to exceed 60 by year-end; both fall inside the band. ### HVAC add-ons (recorded 2026-11-11) Input: Find founder-owned US commercial HVAC services companies with 50-500 employees that could be add-on acquisitions for a PE-backed platform Run record: high effort, 7.6 minutes, 27 pages cited. - #1 Tate Engineering Systems, Inc. (tate.com), Baltimore, Maryland, USA, 146 associates (ACHR NEWS, July 2026); 157 employees (Baltimore Business Journal, May 13, 2026); ~126 on LinkedIn (LeadIQ), Family-owned, fourth generation (founder William J. Tate, 1924; fourth-generation family member Alec Baker President & CEO since 2018); no institutional investors named on its site or in trade coverage: Best fit in the set: pure commercial/industrial HVAC-mechanical services, mid-band headcount with two dated counts showing growth, founder-family ownership, and a top 2026 industry award. - #2 Mullins Mechanical (mullins-mechanical.com), Carrollton, Georgia, USA, Approximately 254 employees as of October 2026 (LeadIQ); the Inc. 5000 profile self-reports company size 501-1,000 - the two published counts conflict, Founder-owned: the company's about page names 'Jacob Mullins, Chief Executive Officer & Founder'; no institutional investors are named on its site: Large founder-owned commercial/industrial mechanical contractor at the top of the band; Inc. self-reports 501-1,000, so headcount sits at the band boundary and needs diligence. - #3 Classic Refrigeration (classicrefrigeration.com), Brea, California, USA, Nearly 200 employees and more than 150 service trucks (own site, 2026); Inc. 5000 company size band 201-500, Family-owned and operated, founded by Danny Lamping (VP/COO) with Dave Rogers (third-generation refrigeration professional) on the leadership team; no institutional investors named: Fast-growing family-owned commercial/industrial refrigeration-HVAC services firm squarely in band, though refrigeration-led rather than pure HVAC. - #4 Accurate Heating, Cooling & Plumbing (accuratehvac.com), Chillicothe, Ohio, USA, Over 150 employees (Chillicothe Gazette, June 20, 2022); ~94 LinkedIn-derived profiles (LeadIQ/index, 2026), Founder-owned and family-led: founder Thomas White is Chairman of the Board, son Brian White is President, Michael Linton is co-owner/CFO; locally owned and veteran-owned; no institutional investors named: Founder-succession moment (founder transitioning to son) plus ACCA commercial-contractor recognition; headcount sources conflict, which caps the rank. - #5 Atlas Mechanical, Inc. (atlasmechanical.com), San Diego, California, USA (Las Vegas office), 95 employees (RocketReach, 2026), Founder-owned: Sam Shekhter, Founder and CEO, leads the company he founded in 1991; privately held with no institutional investors named on its site or leadership pages: Pure commercial/industrial HVAC contractor, founder-led, verified mid-band count; thinner dated momentum and a single dated headcount source. - #6 Jarrell Contracting (Charles E. Jarrell Contracting Company, Inc.) (jarrellcontracting.com), Earth City, Missouri, USA (new Hazelwood, MO headquarters announced 2026), ~99 employees (LinkedIn-derived, November 2026), Family-owned (Jarrell family, 70+ years): President Michael Jarrell and EVP Kimberly Jarrell; no institutional investors named: Clear founder-family ownership and a large dated capacity-expansion signal; headcount is verified only via a LinkedIn-derived count while its own site does not publish a figure. - #7 McCloskey Mechanical Contractors, Inc. (mccloskeymechanical.com), Blackwood, New Jersey, USA, About 96 employees (LeadIQ, November 2026); LinkedIn shows 97 employees, Family owned & operated: 'the McCloskey family has built a strong foundation' over 39+ years; no institutional investors named: Squarely commercial/industrial and family-owned with verified mid-band headcount, but the most recent dated traction signals are 2025, not 2026. - #8 Hoffmann Brothers (hoffmannbros.com), St. Louis, Missouri, USA (also Nashville, Tennessee), Over 300 dedicated professionals (own site, 2026), Founder-owned and family-run: Robert Hoffmann is 'the original owner and founder' (took over Linton Heating & Cooling in 1988); family members Chris Hoffmann (CEO) and Joe Hoffmann (GM/VP) run the company; no institutional investors named: In band and founder-owned with strong growth, but the book is residential-led with a commercial division, weakening the pure commercial-HVAC-services fit. - #9 Lawson Air Conditioning & Plumbing, Inc. (lawsonac.com), Gainesville, Georgia, USA, 100+ team members (HVAC Insider, October 2026); LinkedIn self-declared size range 51-200, with the company description stating it 'employs 100 people', Third-generation family-owned: founder L.D. Lawson (1946); CFO and owner Debbie Lawson Davis; VP Travis Davis; no institutional investors named: Family-owned with a dated capacity-expansion signal and a commercial HVAC division, but the business also carries substantial residential/plumbing work. Search notes: Screened award rosters (ACHR NEWS 2026 awards, ENR 2026 Specialty Contractor of the Year), the 2026 Inc. 5000 HVAC coverage, SMACNews family-business features, HVAC Insider trade press, local business journals, a 2026 HVAC PE roll-up tracker and a 2025-26 HVAC M&A log, then verified founder/family ownership, US HQ and a dated 50-500 employee headcount for each name from its own site, LinkedIn/LeadIQ/RocketReach/PitchBook records and news pages. Nine companies qualify. Excluded for failing the 50-500 band or unverifiable criteria: F.E. Moran (~1,550), John E. Green (1,000+), RK Industries (platform-scale), United Mechanical (PitchBook status: acquired/merged), Old Dominion Mechanical (61 employees, private with no backing, but founder ownership not verifiable), Geauga Mechanical (~37), Willis Mechanical (33), Western Sheet Metal (32), Sav Mor Mechanical (28 now vs 130 in 2022), A&P Air Conditioning (24), The Cooling Company (20), LCS Heating & Cooling (20), Mohr Air (12), Slavik & Co. (8), Anchor Heating & Air (7), C.R. Wolfe (11-50), J.W. Poole (11-50), Carl's Quality (~40), and residential-led shops (Watkins ~60, Parham ~50, Bonfe ~275). The PE tracker and M&A log list none of the nine qualifying names as acquired. Conflicts flagged: Mullins Mechanical (LeadIQ ~254 vs Inc. self-report 501-1,000) and Accurate (150+ in 2022 vs ~94 LinkedIn-derived 2026). ### AI infrastructure seed (recorded 2026-11-11) Input: Find AI infrastructure startups founded since 2024 that have raised a seed round Run record: high effort, 5.6 minutes, 16 pages cited. - #1 RadixArk (radixark.ai), Palo Alto, California, US, null, VC-backed (Accel, Spark Capital, NVentures, Walden Catalyst Ventures, HOF Capital, LDV Partners, AMD, MediaTek plus angels): Cleanest fit: brand-new AI-inference infrastructure company (launched May 2026) with an exceptionally large $100M seed, top-tier investors, and rare pre-revenue traction via SGLang's de-facto-standard open-source adoption - #2 Unconventional AI (unconv.ai), Mountain View, California, US, 61, VC-backed (a16z, Lightspeed, Sequoia, Lux Capital, DCVC, Future Ventures, Databricks; angel Jeff Bezos; co-founder CEO Naveen Rao personally invested $10M): One of the largest seed rounds ever in AI compute infrastructure, led by first-tier investors around a proven repeat founder; founding year rests on TechCrunch's 'one-year-old' description (Dec 2025) rather than a formal incorporation date - #3 Starcloud (starcloud.com), Redmond, Washington, US, 15, VC-backed (Y Combinator, Andreessen Horowitz, NFX, Balderton Capital, Benchmark, EQT Ventures, Sequoia and a16z scout funds, In-Q-Tel): Founded January 2024, seed raised on schedule, and already a $1.1B-valuation Series A company with two satellites launched — strongest executed traction on the list, though it has already graduated beyond seed stage - #4 Upscale AI (upscaleai.com), Palo Alto, California, US, Over 100 technologists on the team, VC-backed (Mayfield, Maverick Silicon, Qualcomm Ventures et al.); incubated by Auradine: New independent AI-networking infrastructure company launched from stealth in Sept 2025 with a $100M+ seed and 100+ team; legal founding year not published (incubated inside Auradine), so the since-2024 criterion is met by its 2025 launch rather than a stated incorporation date - #5 hosted.ai (HostedAI, Inc.) (hosted.ai), San Jose, California, US, null, VC-backed (Creandum, Repeat VC, People Ventures, Z21 Ventures, Golden Sparrow, Hersir Ventures, Tekton); founder-led by Ditlev Bredahl (CEO) with Narendar Shankar, Julian Chesterfield, James Withall: Fits both hard criteria with the company's own published statement ('founded in 2024, launched publicly in 2025') and a closed seed round from a top-tier lead (Creandum); earlier-stage and smaller round than the mega-seeds above - #6 Liquid Compute (liquidcompute.com), New York, US, null, VC-backed (FirstMark, Chemistry, Y Combinator et al.); founded by Ronit Jain (CEO) and Aarav Patel: Founded 2024 (YC W24 batch) with named trading partners and a live regulatory path — concrete traction — but the 'founded 2024' year is inferred from its YC W24 batch rather than a stated incorporation date - #7 Expanse (Expanse Compute, Inc.) (expanse.sh), San Francisco and London; founded in Edinburgh, UK, null, VC-backed (Crane Venture Partners, PXN Ventures, angels); received company formation support and first finance from Edinburgh Innovations; Y Combinator: Textbook fit — founded October 2025 by University of Edinburgh graduates, seed led by a specialist infrastructure investor (Crane Venture Partners), with quantified customer traction ($8M idle compute found in one cluster in a month) - #8 Querio (querio.ai), London, United Kingdom, Team of 8 (1-10 band), VC-backed (BMW i Ventures, LAUNCH, Araya Ventures, Forum Ventures, Haatch, Secways); founders Rami Abi Habib (CEO) and Javier Bonilla (CTO): Founded 2024 with named paying customers and a fresh seed; weakest infrastructure fit on the list (data/BI analytics layer rather than compute infrastructure) but classified under AI Infrastructure by the funding directory and self-describing as AI-ready data infrastructure ## 6. Questions ### What is an AI deal sourcing agent? A research agent that takes an investment thesis and returns private companies that fit it. Nimble's agent searches where private companies show up, such as industry association member lists, trade-show exhibitor lists, registries, fastest-growing lists, local business press and the companies' own sites, then ranks the matches and gives the evidence for each criterion. ### How does it tell who owns a company? From the company’s news and from investors’ portfolio pages. PE ownership is often visible only on the PE firm’s site, so the agent checks there. Subsidiaries that look independent on their own site are caught on the legal or imprint page. ### Does it relax the thesis to fill the list? No. A criterion that cannot be verified means the company does not qualify. The search notes say what was searched and what was left out, so a short list is a real answer. ### Can I run deal sourcing from Claude, ChatGPT or my own agent? Yes. Paste this prompt into Claude, ChatGPT, Codex, Cursor or Grok: "Be my deal sourcing assistant using Nimble. Follow https://docs.nimbleway.com/skills/deal-sourcing/SKILL.md." The skill asks what you need, connects Nimble if it isn't connected yet, and runs the agent. Developers can call the same ready-made agent by its ID from the API, the Python or TypeScript SDK, or from Mastra, LangChain, the Vercel AI SDK and LlamaIndex. ## 7. Skill The skill the prompt points to, as published at https://docs.nimbleway.com/skills/deal-sourcing/SKILL.md (copied 2026-11-11). ### Deal sourcing Builds a list of private companies that fit an investment thesis (sector, geography, size, ownership), with evidence for each criterion. #### What the user gets A list of companies that fit the thesis, with ownership and evidence for each criterion. A short list is an honest answer. #### Agent template This skill runs the Nimble gallery template `deal-sourcing`. The agent's instructions, sources and output schema live in the template. Do not rewrite them in the run. Effort: this skill runs the template at medium effort (about 1-3 minutes), even when the template's default effort is different. That is on purpose. If the user asks for a specific effort (low, medium, high or x-high), run that one instead. #### Flow 1. Interview the user. 2. Connect Nimble if it is not connected already. 3. Run the agent and show the result. 4. Offer next steps. Ask every question with the host's question UI, never as plain text: `AskUserQuestion` in Claude Code, ChatGPT's interactive question form (clickable choices and a Submit button) in ChatGPT. ##### Get the inputs | Input | Required | Ask for | If skipped | | - | - | - | - | | Sector | yes | What the companies do, for example the industry or product | - | | Geography | yes | Country, region or state | - | | Ownership | yes | For example founder-owned, family-owned or PE-backed | - | | Size | no | Employee or revenue range | any size | | Exclusions | no | For example "no PE-backed", or names or domains to leave out | no exclusions | * Skip anything the user already said in this conversation or in their request, and ask only for what is missing. Do not fill in answers from earlier conversations, with one exception: a stable fact about the user, such as what they sell, may be offered as the first option, labeled "(from before)". Never carry over the subject of the request, dates or rules from an earlier conversation. * Ask the required inputs first, in table order, one question at a time. Then ask the optional ones together, in one question-tool call, as many as the tool takes per call (a second call for the rest). * Put the whole question in the question tool, once. Never restate or introduce a question in text. Between questions, acknowledge the answer in a few words, and before the first question, at most one short sentence that acknowledges your role (the user's, if they named one), such as "Sure, I'll use Nimble to be your deal sourcing assistant." Talk in plain, friendly words: never say "input", "required" or "optional", and use no field names. Use plain text only if the host has no question tool, in one short line. * Write 2-3 likely answers as options, based on what the user has already said. Fixed examples in the table are a guide to what the answer contains, not a list to copy. Offer only answers Nimble can act on with public web data, never ones that need the user's own data, such as their customers or CRM. * If you cannot guess likely answers, such as a company name or an industry, still use the tool and offer two or three examples as options, written as answers the user could really give. If the user picks one, use it as their answer. * Every question must let the user type their own answer, optional ones too. Claude's question UI adds its own free-text choice ("Something else" in Claude, "Other" in Claude Code), so add none there. ChatGPT's form does not, so add "Other" as the last option of every question, even one whose options look complete, opening a text field with a placeholder. In any other host, add "Other" unless you know its UI already has a free-text choice. Keep Skip and Other, when you add them, even if you must drop an answer to fit the option limit. * Use single-select only. For a range, such as a size or a date, offer a few options that do not overlap. * Optional questions can be skipped, and a skipped answer uses the "If skipped" value. Claude's question UI has its own Skip button, so add no Skip option there. In other hosts, include a Skip option that states the default in its label, for example "Skip (any location)", not in the question. When you acknowledge a skipped question, say the default you will use, for example "Any location.", not that it was skipped. Do not offer a "skip all" option. If the user asks to skip the remaining questions, use the "If skipped" values for all of them. * Do not ask about effort. Build the agent input from this pattern: ``` Find companies in with . Exclude . ``` Example: `Find bootstrapped European vertical SaaS companies for logistics or freight forwarding with 20-200 employees` * Turn each criterion into a clear test. Ownership is the hard one, so state it exactly. * Do not ask for a number of companies. Narrow theses return few rows on purpose. * Leave out any part of the pattern the user skipped: leaving it out is the default. ##### Connect Nimble Do this after the interview. Until the questions are done, say nothing about connecting Nimble. Check the connection. If a Nimble tool is listed in this session, call `nimble_agents_list`. If the CLI is available, run `nimble --client-source nimble-agent-skills agents list`. A normal response from either, even an empty list, means the user is connected: say nothing and go on. If neither check succeeds (no tool is listed and the CLI is missing or fails, or a check returns an auth error or an authorization link), connect it in these steps. Your message is only these steps: no summary of what the user told you, no mention of the skill, the template or the workflow, and no plan for what happens after connecting (that belongs in the run notice). 1. Tell the user, in two short sentences at most: "Nimble searches the live web and links where every answer came from. Connecting takes one approval, and if you're new to Nimble you start with a free trial." 2. Connect it in the first of these ways that works, and stop at the first one that does: * Show the host's card to connect or install Nimble from nimbleway.com: the Connect card for the connector in Claude, the install card for the plugin in ChatGPT. * If you can't show a card and a check returned an authorization link, show that link to the user. * If you can do neither but can run commands, follow [https://docs.nimbleway.com/agent-onboarding.md](https://docs.nimbleway.com/agent-onboarding.md) and involve the user only for the sign-up and the approval. With only the CLI and no `NIMBLE_API_KEY`, an approval is not enough: follow the page's API key fallback and have the user set the key in their environment, never in the chat. That page is written for you: never link it or its steps to the user. * If none of these works, tell the user how to find Nimble in this app, with the clicks: in ChatGPT, open Plugins in the sidebar, search for Nimble, pick the one described as "Expert web search for agents" and install it; in Claude, open Customize, then Connectors, then Nimble, then Connect. Never send the user to nimbleway.com for this. 3. Ask with the question tool, with the options "I've connected Nimble" and "Not now". If the host has no question tool, end the message by asking them to reply "connected", and add nothing after that. Never ask the user to paste an API key into the chat. When they confirm, check the connection again and go on without comment. If it is still not connected, say so in one line and ask again. If they pick "Not now", say the skill needs Nimble and stop. Show no research until Nimble is connected, not from your own search and not from memory. ##### Run the agent Do not comment on the connection check or on creating or reusing the agent. The first thing the user hears is step 2. 1. Create the agent from the template: `nimble_agents_create(template="deal-sourcing", agent_name="deal-sourcing")`. A 409 "already exists" means it was created before, so go on. If the template is not found, tell the user it is unavailable and stop. Do not substitute another template. 2. Tell the user once, in one line, what you are researching and how long it takes at the effort you will use: "Researching \. Expected to take \