RAG playground
A BM25 ranking with the arithmetic left visible: chunk map, term stats and per-term score contributions.
/rag
Simulation · deterministic model
AI research agent simulator
Ask a research question, then turn the dials a deep-research agent actually has — breadth, source limits, verification, reranking — and watch a deterministic model of what each one costs you.
Research agent simulator
Say it the way you would say it to a person. Your words are matched to one of 18 tools by readable rules, and the tool runs right here in your browser. No upload, no account.
This tool runs in your browser and needs JavaScript turned on.
esc
How to use the research agent simulator
- Type a research question, or start from the example one.
- Set the controls in words — “simulate a research agent with breadth 6”, “simulate it without verification” — or open the settings form.
- Read the modelled coverage, confidence, cost and time, and the queries that breadth would issue.
- Change one setting at a time and compare the runs side by side.
Things you can type: simulate a research agent with breadth 6 simulate verification and reranking
What a research agent is really deciding
A "deep research" run looks like one button, and underneath it is a set of tradeoffs somebody chose on your behalf. How many queries to issue. How many sources to open. Whether to verify a claim against a second source. Whether to rerank the retrieved passages before answering.
Every one of those choices buys coverage or confidence and pays in time and tokens. Seeing the shape of that exchange is the difference between using an agent and being surprised by one.
What the simulation shows
Set a question and the controls, and the model reports:
- The query set the breadth setting would produce, expanded from your question.
- Modelled coverage and the number of sources consulted.
- Modelled wall-clock time and token cost, and how verification and reranking change both.
- A confidence figure that rises with verification and falls as breadth outruns depth.
It is a model, and it says so
No searches are performed and no model is called. This is a deterministic simulation, clearly labelled as one, whose value is the shape of the tradeoff rather than a prediction of a specific run.
Same inputs, same outputs, every time — which is exactly what you want when the point is to compare two configurations rather than to research a topic.
The real thing, when it is ready
FlexContext is building the agent this simulates: research runs over your own library and the open web, with sources kept, claims verified against each other, and every finding cited back to where it came from. The controls above are the ones it will expose, because hiding them is how you end up with a confident answer built on four pages.
Questions
Research agent simulator: common questions
Does this actually search the web?
No. It is a labelled simulation that models the tradeoffs between breadth, verification, reranking, time and cost. Nothing is fetched and no model is called.
What is it useful for, then?
Building intuition about a category of tool that mostly hides its own settings — and for deciding what to ask of a real agent before you pay for a run.
Why is more breadth not always better?
Because breadth without depth spreads a fixed budget across more sources, so each is read less carefully. The model reflects that: confidence peaks and then declines as breadth grows past what verification can keep up with.