Your day. Your market. Your next move.
Broday is market intelligence for real-estate brokers and agencies. A deterministic core turns listings, recorded transactions and building registers into one verified model of the market. A guarded AI layer explains it in plain language.
Live counts from the production database, 15 October 2026. Hover the bars for each year.
The hard part is not generating text. It is knowing what is true.
A broker's market lives in a dozen places at once: listings on several portals, real sales in state registers, buildings in cadastre, clients in messengers. Before anything can be explained, these facts have to be reconciled.
One flat, many identities
The same apartment appears on several portals at different prices. Sales sit separately, under a cadastral number instead of an address.
A price without a reference
An asking price says little until it is compared with what the building actually sold for, on enough evidence to say it out loud.
Opportunities get lost
A fitting object appears, and the client who wanted it is forgotten. Value exists only when the signal reaches the broker in time.
Everything flows into one core. Out comes the next move.
Public registers, market listings and the broker's own clients go in. The core resolves identity, checks every number and produces concrete actions. Every day, without anyone running it.
- One object behind every duplicate. The same flat on three portals becomes one object, by a strict key.
- Building identity. Address and cadastral number tie each listing and each sale to a real building.
- Price compared twice. Against recorded sales in the building or district, and against asking prices.
- A reason in numbers. How much cheaper, against what, on how much evidence. Silence when evidence is thin.
- Listing traps caught. Offices posing as flats, developer stock, multi-unit lots, read from the text by the AI guard.
- Freshness. Stale and withdrawn listings recognised; history never lost.
- 168 data-quality checks. Every day, before anything reaches the broker.
- Requests re-checked every 10 minutes. Every client request against the whole market.
Diagram of the system, not a product screen.
The core decides. The model explains, and is checked.
Broday is not a wrapper around a chat model. The language model sits on top of verified results, sees only what it needs and cannot put a number on screen that the core did not compute.
How one explanation is made
- The core computesPrice reference, evidence depth, match fit, computed before any model is involved.
- A whitelist projectionOnly allowed fields leave the core. Client names never reach the model.
- The model writesOffline, in the daily run, never on the request path, so screens never wait for it.
- Every number is verifiedA text with a figure that is not in the facts is rejected and logged.
- A text twin always existsIf the provider fails or is switched off, a plain template from the same facts is shown. The product never stops.
Live in production today
- AI guard over the top of the rankingReads listing text and flags traps with a quoted piece of evidence; 580+ listings judged so far. A word list reached 24% precision on the same task; the guard reads meaning.
- 4,900+ model-written explanationsWhy this object, what changed, morning brief, match and digest. District and next-move texts run on templates for now.
- 1,100+ texts rejected by the checksEach one replaced by its template twin and logged with the reason. Nothing unverified reached a screen.
- Two providers with automatic failoverProvider-agnostic transport; a failing key moves work to the second provider and back.
- Every call meteredTokens and cost recorded per call against a price table, so unit economics are measured, not guessed.
Explain what the core knows
Briefs, reasons and summaries from computed facts. Cheap per broker, cached, with a template twin.
Requests in plain language
"Two rooms, quiet street, within budget" becomes validated request criteria, by text or by voice, confirmed by the broker.
Prepared actions
Follow-ups, client messages and reports drafted by the model, sent only after the broker approves them.
Built so that every number can be trusted.
These are signals of engineering maturity, not of market traction. They show that the system is ready to carry real users.
Every number is provable
Each figure has a source and a way to recompute it. Where evidence is thin, the system stays silent and says why.
Client data is encrypted
Personal data is stored encrypted, and every time it is shown is written to an access journal.
Workspaces are isolated
One agency never sees another's clients. A table that could leak across workspaces turns the build red.
Runs by itself
The daily market run is automatic, catches up if it misses a slot, and a missing run is visible without anyone checking.
Watched around the clock
An outside monitor checks the system continuously, and no signal counts as a failure. Only a fully green build reaches production.
A market is a package
Sources, geography, thresholds and language live in a market package, not in code. A blocking build check keeps the core market-free.
Brokers first. Then the whole agency.
Brokers work their own clients; the agency sees the whole picture and keeps its knowledge when people change.
Broker and agency workspace
- 1Roles checked by the serverBroker and agency administrator, each with exactly the access they need.
- 2A shared agency baseThe administrator sees all clients and deals; each broker sees their own.
- 3Coordinated seller workColleagues' actions on a seller lead are visible, so nobody calls twice.
- 4Knowledge stays with the agencyWhen a broker leaves, their clients stay.
Where the platform goes
- →Team tasks and hand-overAssignees and passing a case to a colleague.
- →Selections for clientsSeveral objects in one link, and a daily selection.
- →Publishing for agenciesAgencies publishing their own objects on Broday.
- →A broker networkBrokers exchanging supply and demand with each other.
Engineering first. Market proof next.
Broday is pre-revenue. The product runs in production on live market data; the next milestone is repeated use by real brokers. Every step ends with something measurable.
Working core
Data foundation, decision engines, broker workspace and Level 1 AI in production.
Broker pilot
A small group of brokers on the full product, with explanations measured for usefulness and cost.
Plain-language requests
Level 2: text and voice into validated criteria, compared against the classic workflow.
Second market package
The same core with a new package of sources, geography and language.
What we are looking for.
Grant programmes, model providers and investors who want to back a working system rather than a slide deck.
What support pays for
- API credits for the broker pilot Level 1 and 2 for every pilot broker, with cost measured per call.
- Model and evaluation partnership grounded multilingual output, refusal on thin evidence, safety of prepared actions.
- Funding for the pilot and the second market people on the ground, data access and a second package.
What a partner gets
- A real domain with hard data identity resolution and evidence, not toy prompts.
- Inspectable seams source → computation → projection → model → check → fallback.
- Honest measurement usefulness, cost and fallback rate reported from production logs.
Broday in one table.
- Product
- Market intelligence and daily workspace for real-estate brokers and agencies.
- Category
- Decision support and market analytics. Not a listing portal and not a CRM replacement.
- Stage
- In production on live market data · pre-revenue · first broker pilot next.
- Data
- 334,000+ recorded transactions since 1998, 24,900+ buildings, 23,500+ listings, energy certificates, cadastral values and construction files.
- Core
- Deterministic: identity resolution, price references, evidence depth, matching, 168 data-quality checks a day.
- AI
- Live: AI guard over the ranking, 4,900+ model-written explanations, 1,100+ rejected by number checks. Whitelisted input, number verification, template fallback, two providers, metered cost.
- Engineering
- 11,800+ automated tests, 127 nightly browser scenarios, 30-minute rebuild from an empty machine, encrypted personal data with an access journal.
- Platform
- Market-specific rules live in a package; the core carries none. A second market is a package, not a rewrite.
- Looking for
- API credits, model partnership, grant or investment for the pilot and the second market.
- Contact
- hello@broday.io
See it working.
A live walkthrough of the product on real market data, for programmes, partners and investors.
hello@broday.io