Analyze mode, on the web

An analysis that shows its work.

Ask a question about your data or the world's. One AI writes and runs the code, teammates you approve can take parts of the job, a second AI checks the result, and every figure is traced to where it came from. You get back an answer, a report, a model or a diagram, and you keep working with it.

Open Analyze Analyze FAQ

symposiumai.app/analyze

A team run from start to finish: the prompt, the plan you approve, teammates working, and the finished report.

01How a run works

Pick the roles, then ask.

Choose an Operator to do the work and a Reviewer to check it, add models to a team pool if you want help, and start from your own question or one of the example prompts.

The Analyze composer with role chips for Operator, Reviewer and Pool, Team mode on, and the What you get back menu open — The composer: Claude as Operator, ChatGPT as Reviewer, Gemini and Mistral in the team pool, and the menu of what you get back.

01You ask, with your data or ours: Upload files (CSV, Excel, JSON, PDF, images, up to 100 MB each), pick data connectors, choose what you want back or leave it on Auto, and ask in plain language.
02The Operator does the work: One AI runs a real tool-use loop: it writes and runs Python in a private sandbox, searches the web, fetches data, and keeps going from its own results until the job is done.
03It keeps running without you: Analysis runs on our servers, so you can close the tab or the laptop and come back to a finished result. Run several at once, each in its own tab; the bell tells you when one needs you or is done.
04A second AI checks it: The Reviewer reads every report and returns prioritized points. You choose which go back to the Operator.

02Team mode

A team you approve before it starts.

Turn on Team mode and the Operator proposes a plan: an independent panel, where several models work the same question and the Operator compares their answers, or a split, where each takes a different part. Nothing runs until you approve it.

A team plan card: Claude proposes a split of four tasks across Gemini and Mistral, with Approve plan, Decline and Have Claude do it buttons — The plan. What each teammate will do and what it returns, with the cost shown. Swap models, add one to a task, decline with feedback, or have the Operator do it alone.

The Team panel with four tasks running, each showing its model, status, time and tool count — The team at work. Each teammate's progress, tools and time in the Team panel. Stop one, or replace it with another model, while the rest keep going.

One Operator writes the report from its teammates' work. The Reviewer checks it like any other report.

03Review

A second AI checks the work, and can redo it.

The Reviewer reads every report and returns prioritized points, each with its confidence and the files it applies to. It never edits the report.

Check independently sends a point to a different AI that re-fetches the sources and redoes the numbers, then marks it Confirmed, Disputed or Unverifiable with its evidence.
Nothing changes until you choose which points go back to the Operator.

The Review Queue with a prioritized point from ChatGPT, its confidence and files, and the Check independently button — The review queue on a team report. In this run, ChatGPT independently re-fetched the World Bank data behind the income figures and confirmed them.

04What you get back

Outputs you keep working with.

Pick one, or leave it on Auto and the AI picks from your question. Reports, models and diagrams are kept in your Library.

Answer

A reply in the chat, with no file unless you ask for one.

Report

Findings to read, share and export: PDF, PowerPoint, Word or an interactive page, plus an Excel workbook of the data behind it.

Model

Inputs you change and run again, with results that update. Name scenarios, compare any two, or download it to run on your own computer.

Diagram

A data model or a process flow you edit by hand or by asking. Every save is a version you can go back to.

Simulation

A process run many times to show where work waits and what more staff would do, as a model whose scenarios you compare.

A report's Sources page listing World Bank and UN population sources with the request URL and the fetched file — A report with its sources. Every sourced figure carries a marker; the Sources page lists the request, the dataset and the file behind it.

symposiumai.app/analyze

A model you run again. A mortgage model built on today's FRED rate: change the down payment, run it, and compare the two scenarios side by side.

05Primary sources

38 data connectors, built in.

The Operator fetches from the source, not from a search snippet, and every fetch records the request, the time and a fingerprint of the response.

Weather: Weather.gov (NOAA), OpenWeatherMap
Finance: FRED (Federal Reserve), SEC EDGAR, Yahoo Finance
Science: USGS Earthquake Data, arXiv, Semantic Scholar, PubMed / NIH
Reference: Wikipedia, Wikidata, Open Library, Dictionary API
Academic: Crossref, OpenAlex
Health: ClinicalTrials.gov, openFDA, WHO Global Health Observatory
Government: US Census Bureau, NASA Open APIs, EPA Envirofacts, FBI Crime Data (UCR), World Bank, Socrata / data.gov
Economics: Bureau of Labor Statistics, OECD SDMX, IMF DataMapper, Eurostat, UN World Population Prospects
Developer: npm Registry, Hacker News, PyPI, Stack Exchange, Libraries.io, GitHub
News: NewsAPI
Geography: OpenStreetMap / Overpass
Data: Google Sheets / CSV

06On our servers

Several analyses at once.

Each analysis runs in its own tab and keeps going when you close it. The bell says when a plan needs your approval and when an analysis is done.

The Analyze tab strip with several sessions and the notifications panel listing finished analyses and a team plan that needs approval — Session tabs across the top and the notification bell open: two analyses finished and a team plan waiting.

07Your data

Where your files go.

Files you upload, data the AI fetches, the code it writes and the files it produces run in a private, isolated sandbox for your session, hosted by our code-execution provider E2B, with no internet access.
Uploads are also kept with the session in our cloud storage, so they are restored if the sandbox is recycled. Sandboxes are deleted after 7 days of inactivity, when you delete the session, or when you delete your account.
Your provider keys are stored AES-256 encrypted on our servers and never sent to your browser. You pay each provider directly, at their prices, with no markup.

Analyze needs a provider with function calling: Claude, ChatGPT, Gemini, Grok, Mistral, Cohere, Kimi, or GLM. The details are in the privacy policy.

Start from an example, or from your own data.

Open Analyze