It reads the whole record, not the page you happen to be on
Ask a question in plain words and it goes and gets what it needs: the award history, the open notices, the subawards, the vehicles, the buying offices, the vendors, the rules, the bills and everything else on file, up to six searches a question, across millions of records back to FY2001. Then it answers, with every figure counted over the whole matching set, and tells you which searches it ran so you can open any of them and check it.
Ten places it works, and what each one answers
Not one chat box bolted to the side. A reading sits on the screen it belongs to, shaped for the decision that screen exists to support.
A brief of any open notice
What they need, who is buying, the dates, the set-aside and who can bid, a fit score from 1 to 5 with the reason, what happened last time this buyer bought it, what to do next, and the questions to ask the contracting officer.
A reading of any rule
What it changes in a supplier's terms rather than the agency's, which kind of firm carries the cost, what the market behind the tagged codes is worth, and whether to comment and by when.
A reading of any bill
What it does to federal buying, how far from law it actually is given the stage it reached, and what it would mean in the codes it names. It never predicts a vote.
How a company wins
Any competitor's award history read as a competitor would read it: what they really do, who buys it, whether they win on set-asides or on price, and whether the record shows an opening.
How an office buys
What a buying office buys, how it buys it, whether the same names keep winning, what is open now, what ends soon, and what the record actually suggests about getting in.
What a budget line means
What a line of the defense request pays for, which way the money moves across the years the request states, and who would sell into it.
What a market is like
Any NAICS code: who buys it, what a realistic first win looks like against the median, how crowded it is, and what to watch in the next year.
Whether a vehicle is worth chasing
Single or multiple award, how much of the ceiling is gone, how long is left, and whether the record shows a way in or none at all.
What an award tells you
What was bought, how it was bought, what the competition and offers say about how hard it would be to take, and what to be doing before it ends.
A question, worked across everything
Ask it the way you would say it. It runs its own searches across every list in the product, up to six for one question, and answers with totals counted over all of it. It names every search it ran, and each one opens as a screen you can check. You can also ask a question about the screen you are on, from any screen. The app sends where you are and what you asked, the facts are read again on our servers from the same queries that drew the screen, and the answer is written from those. It works on your own bids, contracts, pipeline and books as well as on the record.
It can reach everything, and it may only use what it found
It searches the record itself, through the same queries the screens use and under your own plan limits, so it can only reach what you could have found by clicking. Every search it runs answers with the total and the money over the whole matching set, the breakdown by code, buyer, state and year, and a sample of records to name names with. That is why a figure in an answer is a figure over millions of rows rather than over twenty lines.
What has not changed is the grounding. It may use only what its searches returned. It cannot recall a number from training, it is told that anything the record does not show does not exist, and where the record does not settle a point it has to say so rather than fill the gap. Every answer lists the searches behind it with what each one found, and each one opens as a screen, so any figure can be checked in a click.
- Every search runs server side, through the same queries the screens use
- Totals are counted over the whole matching set, never over the sample
- Every answer names the searches behind it, and each one opens as a screen
UEI, CAGE, then normalized name
Every new notice tagged by work type, process and material
This one runs on the way in rather than on request. New notices are classified overnight, watched codes first: machined parts, sheet metal, welding, coatings, heat treating, electronics, fasteners, fluid, bearings, seals, plastics, tooling, repair, engineering, raw material, IT and software, professional support, construction, facilities, logistics, medical, training, research, security or supplies. Plus the processes named in the text, the materials or systems, a quantity hint, and a supplier fit score.
The point of doing it on the way in is that your filters then work without you writing keywords. The tags become filters and the fit score sorts Today.
Re-reading something costs you nothing
Your plan includes a number of new readings each month and the app shows what is left. Opening a reading that already exists does not count again, so a record you come back to, or one a colleague has already opened, costs you nothing to read again. When the month's readings are gone the app says so plainly instead of failing.
If you want more than your plan includes, the owner of an account on a paid plan can add an AI key of their own on the Users and billing screen, from any of nine providers. The owner picks a named model and how long an answer may be, and chooses when the key is used: for every reading, or your plan's readings first and then the key. A reading written on your key is billed by your provider and does not come off the month. The key is tried before it is kept, encrypted before it is stored and never shown again in full, and it can be removed at any time. If it stops working, the app says so and goes back to the plan's own readings. The rest is on Your own AI key.
Where it stops, stated plainly
It can misread. It is a model reading a federal record, not a lawyer, an accountant or a contracting officer. Every AI-written block on every screen is labeled as one and says what it was written from. It is not legal, financial or procurement advice, it does not replace the solicitation, and nothing it writes should be quoted to a customer without checking it against the record printed above it. Your data is never used to train anything, and nothing you type is stored as training material. AI writing in a proposal draft is apart from the ten above, and it is optional. It is the one place a model is sent passages of your own documents, and it is off until the owner of the account or a manager turns it on. The security page says what it sends.
Related
- Open noticesWhere the briefs live.
- Rules and fundingThe layer the readings help most.
- Reading a solicitationWhat a brief cannot replace.
- Data and sourcesWhat the model is reading from.
See who was last awarded in your NAICS
Every code, every buying office, every vendor and every recompete, organized from the federal record and read nightly. Start on one code with no card. A plan opens every code.
Try the live demo The real app on one code. No email needed.