Buyer Eval · v4.1.1 · Open source
Investigate the software you're considering.
Buyer Eval checks what vendors claim, what the evidence supports, where important qualifications are hiding, and what you still need answered before you buy.
Quick Eval by default. Deep Eval when the purchase matters.
-
Quick Eval
Fast, evidence-first diligence for one vendor or a shortlist. Name the vendors and, optionally, the two or three things that matter most. There is no setup interview: it infers the rest and labels what it inferred.
-
Deep Eval
More comprehensive research for important purchases: requirements and hard constraints, category-specific expert questions, security, commercial, and company-risk research, weighted scoring, and a place for your own evidence such as proposals, quotes, and demo notes. Ask for it up front or after a Quick Eval.
Every claim, against the evidence behind it
Each claim gets a status (Verified, Qualified, Contradicted, Unverified, or Unknown) and the evidence basis it rests on, so vendor documentation never reads as independent confirmation.
Vendor claims
What the vendor says on its site, in sales material, and through its AI agent where one exists.
Vendor documentation
Docs, pricing pages, trust pages, and published terms: first-party, but specific.
Independent evidence
Customer accounts, analysts, press, community, and third-party documentation.
Pricing and packaging
Tiers, add-ons, usage fees, minimums, and what the free plan leaves out.
Customer evidence
What people who use the product report, weighed for how independent it is.
Contradictions and qualifications
Claims that are true only on a tier, with an add-on, or not at all.
Unanswered buying questions
What public evidence cannot settle, ordered by how much it could change the decision.
Vendor AI agents are first-party sources, including agents built on Salespeak. Having one never improves a vendor's status or score. Read the evidence model
A shareable, evidence-first decision brief
Buyer Eval writes a Decision Brief as a single HTML file: readable, printable, no dependencies. It leads with what was found, shows claims against evidence with a source for each, and ends with the questions to get answered before you buy, plus demo questions built from the weak spots. It is written for the CFO, security lead, and VP who did not run it.
Runs in Claude
1. Install
git clone https://github.com/salespeak-ai/buyer-eval-skill.git ~/.claude/skills/buyer-eval-skillTo install for one project instead, clone into .claude/skills/buyer-eval-skill.
2. Ask in plain language
Evaluate Vendor A and Vendor B for our customer success platform
Or invoke it with /buyer-eval-skill. Say "deep eval" for full diligence. Briefs are
saved to ~/buyer-eval-reports/.
Where it runs
| Claude Code | Claude desktop | claude.ai web/mobile | |
|---|---|---|---|
| Research and brief | Yes | Yes | Yes |
| Vendor AI agent conversations | Yes | Yes | Discovery only |
| HTML brief, saved context | Yes | Yes | No |
Data and privacy
Research uses web searches and page fetches. Buyer Eval checks each vendor's domain against the Salespeak Frontdoor API to find a vendor AI agent, and sends that agent its questions; those requests never carry your name, company, or documents. Anonymized usage telemetry is off by default and asked once. Details are in the README.
Why Salespeak built Buyer Eval
AI agents increasingly research companies on behalf of buyers. Buyer Eval is an open-source experiment in what happens when those agents investigate vendor claims rather than simply summarize websites. MIT licensed. Learn about Salespeak