Every product here does one kind of work: it takes what an organisation already holds — its AI systems, its suppliers, its paperwork, its own written procedures, what the public web says about it — and turns it into something you can act on and defend. The platforms built for that work start at fifty thousand a year and are quote-only. We build for the businesses and communities below that line.
Bureau's work is read by a person; Conduit's is read by software. That changes everything about how each one is sold, so they are run as separate concerns rather than one catalogue. Commissioned work is built into whichever of the two it belongs in — it is not a third business. Every live product has an open, stateless endpoint you can run against right now — no signup, no key, no sales call — so you can judge the engine before you talk to us.
Software for a named buyer with a job to do — a compliance officer, an operations lead, a back office. Bought deliberately, run for years. Each one refuses to answer past the edge of what it actually knows.
Classifies an AI system by risk and derives the obligations that attach — and tells you which apply today versus which commence later. Every assessment is hash-chained, so a quietly edited history breaks the chain and says where.
A global risk index tells you the world is on fire. Bellwether tells you the one plant you single-source your castings from is in the part of it that's burning. Alerts fire on the crossing, not the level — so it stays unmuted.
Works out who goes where and in what order, respecting who holds which skill and whose day is already full — then writes the follow-up messages a good shop always means to send. Anything that won't fit is reported, never silently dropped.
Reads an unstructured agreement, pulls the fields that matter, and says what's missing or doesn't add up — then generates the summary and closing checklist. Metered per document, priced to the labour displaced rather than per seat.
Your own documentation, made answerable — with a citation to the exact section, and an honest miss when the documents don't cover it. The how-to ones become checklists it walks a person through.
When someone asks an AI to recommend a business like yours, it can only put forward what it can find and verify. Herald measures the five conditions that decide that and says exactly what to fix. Every fix is a change to your own real information — nothing manufactured, nothing posted on your behalf.
Infrastructure sold to machines rather than to people. Discovered by an agent, paid for per call, with no account to open and no one to talk to. Priced by the request, because the thing holding the card is software.
Macroeconomic signals served over the Model Context Protocol and billed per request. Built for agents, not dashboards — an automated consumer pays for the calls it makes, with no account to open first.
Searches independent sources for a claim and reports what each one actually says — supports, contradicts, or silent — with a diversity-weighted agreement score. One wire story in forty outlets is one source, not forty, so syndication stops counting as confirmation.
The plumbing behind charging for an API: usage metering, idempotent billing, self-serve signup and a customer console. Ten text and document endpoints priced per call. The infrastructure is the asset — it turns any endpoint you own into one that can take money.
Pricing scales with actual use: systems monitored, suppliers watched, technicians and users supported, documents processed, or API calls made. The billing unit is stated beside every price.
Planning guidance, not a final commitment. These are recommended entry prices anchored to market comparisons. Final pricing depends on scope, volume, implementation requirements and contract terms; enterprise plans are quoted.
Plans scale by the operating unit each product supports
Usage-based infrastructure billed per API call
Commissioned work carries no list price. A build for one organisation is quoted against its scope, and it is billed inside whichever division it belongs to — as a Bureau product if a person uses it, as Conduit infrastructure if software does. How that works →
Not positioning statements — each one shows up in how the services behave, and you can check every one of them yourself.
Scoring, routing, extraction and retrieval are ordinary computation — same input, same output, with no model call in the path. A language model can improve the presentation; it is never the thing standing between you and an answer.
The expensive failure here isn't a missing feature, it's a confident wrong answer. Playbook refuses off-corpus questions, Circuit reports the job it couldn't place, Clerk marks a field unfound rather than guessing, Attest declines to assess against a repealed statute.
The software drafts, scores and schedules. A person sends, publishes and pays. That line is structural rather than a setting — the products that generate outreach have no capability to deliver it, and none of them can move money.
Where a record matters, it is tamper-evident. Buyers and regulators don't only want your current answer — they want evidence the history wasn't rewritten afterwards.
For the common good. It is on the arms because it is the test the work has to pass: software that serves the people who could never buy the enterprise version.
Everything above began as one organisation's specific problem. If yours isn't on the list, that isn't a dead end — it's the more interesting conversation. Commissioned work isn't a side business here: it is how the catalogue gets written, and a build lands in Bureau or in Conduit depending on whether a person or a machine will be reading its answers.
One person builds your software and stays with it afterwards. You'll know who wrote it and be able to ask them why. No account manager relaying questions to a team you never meet.
The commitments on this page aren't reserved for the products. Deterministic where it matters, honest when it doesn't know, nothing sent on your behalf, and a history you can check — those hold for work built for one company too.
Start by describing the problem, not a spec. We'll tell you what it would take, what we'd leave out, and — if an existing tool already solves it — that you shouldn't pay us to rebuild it.
The point isn't a handover and a goodbye. Software that goes into daily use needs someone answering for it, so maintenance is part of the arrangement rather than a separate sale.
Buying one of the products? Use the indicative prices above to plan, then tell us roughly what you're working with — AI systems, suppliers, technicians, users, documents or monthly API calls. We'll confirm the right usage tier and a real number, not put you through a discovery-call maze.
Need something built? Describe the problem in plain terms — what the work looks like today, who does it, and where it breaks. That's enough to start. You'll get an honest answer about whether it's worth building, including if the answer is no.
inquiries@forthecommongood.ai