Python 3.11+ · zero dependencies · Apache-2.0

You cannot see what assistants say about your product.
Now you can ask them.

askround asks ChatGPT, Gemini, Claude and Perplexity the same question through their APIs, with web search on, several times each, and reads whether your product is named, at which position, beside which rivals, and what they claim about it. Your coding agent runs it itself over askround mcp.

Install it ★ View on GitHub

What your agent does with it

TaskDo assistants recommend Modelado to someone without an NVIDIA GPU?

askround_doctor257 tokens

can run: openai, gemini (anthropic, perplexity: no key)

askround_ask  question: "What Windows app can turn a single photo into a 3D model without an NVIDIA GPU?"  products: ["Modelado"]  extract: true

304 tokens

Modelado   openai 1/1 #1   gemini 0/1
rivals     TripoSR · Shape Studio · Modly · TRELLIS.2
cost       $0.18   in: openai#1

askround_answer answer: "openai#1" around: "Modelado"175 tokens

“Modelado (Windows desktop app): turns one photo of a single object into a textured mesh (TripoSR under the hood).”

pasting the answers in: 12,253 charactersaskround's brief: 614 characters, each fact cited

Every fact in the brief names the answer it came from, so the agent checks one passage instead of reading the round.

No account, your own keysRefused before it is paid forMeasures the APIs, not the apps

4 assistants, in parallelA brief of ~300 tokens per roundMCP server + Claude Code plugin

A real round

One question, two assistants, two different answers

Asked on 2026-11-12 through the APIs with web search, one answer each, about Modelado, a Windows app by the same author. Recorded, not mocked: these are the answers askround reads in its own tests.

The question

What Windows app can turn a single photo into a 3D model without an NVIDIA GPU (AMD or CPU only)?

ChatGPT (OpenAI API)
gpt-5-2025-08-07
Modelado: Named, #1
What it recommended, in order (as the extraction pass read it, with gpt-5-mini)
1. Modelado (Windows desktop app)
2. TripoSR
3. Shape Studio by Hivemind (free)
4. Modly (open‑source desktop app)
5. TRELLIS.2
6. Hunyuan3D
7. TripoSG
8. Reallusion Character Creator + Headshot 3 (plugin) — faces
9. FaceGen Artist (faces)
10. Blender
11. MeshLab
Gemini (Google API)
gemini-2.5-flash
Modelado: Not named
What it recommended, in order (its own bold headings)
1. Modly
2. RealityScan
3. DreamVision
4. Fudge3d
5. Image-to-3d-model

The round cost $0.18 on the author's keys. Answers vary from run to run, so a real round asks each question several times and reports how many runs name the product. askround measures the APIs, not the chat apps, which add memory and personalisation of their own.

What a round reports

Read the answers as data, not anecdotes

Named, and where

For each product, how many runs name it and its position in each, per provider and per locale. A table's "best fit" row counts as a position.

Rivals and claims

The products named beside yours, and what each answer claims about yours, quoted, with the answer it came from, checked against the facts in your product file.

Pages cited

Which of your pages the assistants cite, and the ones they cite instead. A prompt set runs the same questions every week, and diff says what moved.

For agents

Built for the agent that asks, not the person who pastes

askround is a CLI and an MCP server with zero runtime dependencies, in the family of roadkeep: small answers on purpose, refusals before money is spent, and nothing an agent has to read twice.

A brief, not a transcript

askround_ask returns a few hundred tokens for a whole round: each product's runs and positions, the rivals, the claims, each fact with the ids of the answers it came from. askround_answer returns one answer, or only the passage behind a fact.

Refused before it is paid for

A question that names the product it tests, personal data, a missing key or a round estimated above the cap is refused before the first call, every problem at once, each with a code and its fix.

The person sets the money

A day and a month budget in a config file no tool writes. An agent cannot raise it, and a round that would pass it waits for the person's approval.

Long rounds as jobs

A round takes minutes. Started in the background, it returns a job id at once and keeps every answer as it lands, so an interrupted round loses nothing it paid for.

The same discipline as roadkeep, the roadmap tool these were built with. See roadkeep

Install

Python 3.11 and the API keys you already have

In Claude Code, the plugin: the MCP server, four slash commands and the skill

/plugin marketplace add alegauss/askround
/plugin install askround@askround

Or the command line, from a checkout until the first release

$ pip install .
$ askround doctor
$ askround ask "What Windows app can turn a photo into a 3D model?" --product "Modelado" --extract

Wire a project for everyone who clones it

$ askround install           # names the plugin in .claude/settings.json
$ askround install --vendor  # and a copy for Claude Code on the web

Keys are read from OPENAI_API_KEY, GEMINI_API_KEY, ANTHROPIC_API_KEY and PERPLEXITY_API_KEY. askround never stores them, and doctor names the variables, never their values.

Cost

Measured, in money and in tokens

Every round prints an estimate before it runs and its actual cost after, from a dated price table. What each MCP response costs the agent is measured in the tests on recorded rounds, against a budget per tool that a change cannot pass unnoticed.

MCP response	Tokens	Budget
askround_ask, the brief	304	400
askround_ask with detail	1,360	1,800
askround_answer	868	1,100
askround_answer around a quote	175	250
askround_run, a whole prompt set	760	1,000
askround_diff	112	200

Two providers, three runs, extraction on; one token per four characters of JSON.

Measured on itself

Do the assistants name askround? Not yet.

askround keeps a prompt set about itself, the questions a developer would ask without knowing it exists, and this is its latest round, from 2026-11-12: one run per question on each provider, $0.34 in all. It is shown as it came back.

Question	ChatGPT (OpenAI API)	Gemini (Google API)	What they named instead
How can I track whether ChatGPT and Gemini recommend my app when people ask for a tool like it?	not named (0/1)	not named (0/1)	Reddit, X, App Store analytics, GA4, Looker/Data Studio
Is there an MCP server that lets my coding agent ask several LLMs the same question with web search?	not named (0/1)	not named (0/1)	LangChain/LangGraph, OpenRouter MCP server, LiteLLM MCP gateway, Microsoft AutoGen, LiteLLM Agent MCP Server
What open-source command-line tool measures a product's visibility in AI assistant answers through their APIs?	not named (0/1)	not named (0/1)	aeo-platform, AVM, Genwolf AI, OneGlanse, Elmo

The set is examples/askround.toml in the repository. examples/askround.toml