B Team (working title)
AI-EMPOWERED PRODUCT DEVELOPMENT

Get the software your NGO needs, without a big engineering team.

AI builds, every output is checked, and the records your compliance needs are made as you work. B Team (working title) is for NGO directors and for the product owners at organisations who need software and have few developers.

Tell us you are interested See how it works

Early build, not offered to anyone yet. NGOs come first. Most of what this site shows is planned, and each part says so. What is not built yet

Overview: where the project stands and what waits on you this turn. The workspace is organised the way a product owner thinks, from what the product is for to what it complies with.
Where a project starts

Say what the product is for before anything is planned

The first page of a project is its vision: what the product is for, who it is for and who it is not for. An agent drafts it from your words, and every agent and reviewer reads it before any work starts. Planned, not built

Vision: what the product is for, who it is for and who it is not for. An agent drafts it from your words; nothing is adopted until you approve it.
Five questions an organisation's product owner asks

Start from your own worry

AI can write code now. That changes what is hard. Each question has a page of its own.

How it works

Decide what matters. Hand it over. Check the result.

As AI writes more code, the hard part moves to deciding what to build, managing the work, controlling the results and testing with users. That is our view, not a finding. Effective comes before efficient.

1

Decide what matters

Write your goal in your own words. The engine drafts one issue with a done-when, and lists the questions it could not answer. You approve it before anything starts. Planned, not built

Today, new work is triaged before anything starts. Measured

A goal becomes an issue. The draft cites the words it rests on, and it does not guess: an unanswered question goes into the issue as an open question.
2

Hand it over

A planner proposes which tasks can run together and which must wait. You can change the plan. A rule in the machinery, the register, refuses a task that would clash with one already running. Then agents do the work. Planned, not built

Today, a coordinator plans a job and the machinery runs one step at a time. Measured

The planner's proposal in three columns, with what each task touches, when it comes back, what it costs, and a clash the register refuses.
3

Check the result

Five standard reviewers and the project's own checks look at the change. You try it, and only you can accept it. Nothing goes live without a person saying yes.

Reviewer and gate today: Measured (pair runs) Measured, with a recorded deviation Five standard reviewers: Planned, not built

Trying the change in a preview first is planned. Planned, not built It needs the sandbox, which is not built (issue #176).

One change, gathered for your decision: what it covers of the done-when, the project's checks, and the buttons that are yours alone.
Who gains what

One promise for each person around the table

  • Director or product owner

    See what is happening and what waits for you, on one page.

    Built

  • Programme lead

    Say what you need in your own words, and judge it by whether people use it.

    Planned, not built

  • Engineer or volunteer developer

    Read what matters most, at a level your team chooses, with the file and line behind each finding.

    Planned, not built

  • Funder or board

    Get a read-only view of what was asked, built, accepted and what it cost.

    Planned, not built For your board or funder
The layer across all five questions

How we raise reliability: no agent is trusted, every output is checked

AI does the writing and part of the checking. It is a worker inside a set of rules. Each rule is traced, tested or checked, and each says what is built and what is not.

  • Traced

    Everything an agent does is recorded, and can be read back.

    Every model call is recorded, and you can read who said what to whom. Measured In plain words, in your own workspace: Planned, not built

    Claims 8, 9 and 71

  • Tested

    No instruction or model change is used before it passes its test set.

    The gate exists; a person runs it. Measured; gate manual The system enforcing it, and open models measured against a top commercial one: Planned, not built

    Claims 20, 60 and 69

  • Checked

    AI writes the checks, the checks run without AI, and a person approves.

    A person's approval, and a separate reviewer: Measured, with a recorded deviation Checks written by AI and run without it: Planned, not built

    Claims 13 and 70

How we raise reliability, with its limits.

Candour

What is in place and what is not yet

We have no customers to show you yet. So here is what we can show: each row names its check, or says it is not built yet.

No. Claim State What it serves Check or issue
10 A separate reviewer checks work before you see it. Nobody has shown that a reviewer catches a real fault. Measured (pair runs) Independent review (F13) reviewed_checks; #28
13 Nothing is published or merged without a person. No self-approval. The recorded deviation: whoever holds a project's key (Temporal token) to the engine's workflow service can send a decision under any name. Measured, with a recorded deviation Human approval (F9, S9) inbox_checks; #32
5 A spending limit is enforced before a call. Measured Cost caps before a call (N3) ledger_checks; test_budget
19 Text from outside the team is data, never orders. Measured Outside text is data (S8, S14) injection_checks (9 attacks)
31 Set up a project from a form that opens a merge request. Built Project setup from a page (#239) test_project_form (stand-in)
29 A page shows running or stopped, all jobs, waiting items, cost. Built Web interface (F17) test_page; test_jobs_page
45 A planner proposes safe parallel work. You can change it. A register stops clashes. Not built Planner and register (F21) #286
46 Five standard reviewers, one per kind of review: security, documentation, compliance, Done when, architecture. Not built Review follows the risk (F22), compliance reviewer (C10) #107; #299
28 A data policy says which provider may see a project's data. Not built Data policy per project (S12), providers and transfers (C4) #57
25 Secrets stay out of prompts, traces and logs. Not built (not met) Secrets (S5) #51, #84, #57
62 Each change has a record: diff, test result, findings, risk note with rollback, approver, time. Not built Change record (C16) #307
66 A processing agreement before B Team sees personal data. No own use. Its access off by default. Not built Processing agreement (C1), no own use (C2), access off by default (C3) #297, #300
67 Independent assurance for B Team in stages, with dates. B Team holds no certificate today. Not built Independent assurance in stages (C29) Waits for a decision by the founder (Codrin). #306
52 Five measures are recorded by the machinery and shown together. Not built The five measures (F27) #290

All 73 claims, with their checks · What is not built yet

Waiting for real proof. These empty boxes are where a pilot partner's logo, a measured figure and a quote will go. Nothing in them is real. They stay empty until a pilot NGO has real figures and agrees to be named.

  • [figure] [hours saved per month]
  • [figure] [share of changes accepted at first review]
  • [figure] [cost per accepted change]
[Quote from a pilot NGO, with name and role]
Frank answers

Questions you would otherwise email us

Which AI models does it use, and can we choose?

The model for each role is a setting, not code, so you can choose. Three providers answer, two are wired to the gateway, and all four roles decided so far run on Anthropic today. A European route through Scaleway is planned and not connected. Choose your models, keep your freedom

Will our data be used to train anyone's models?

We cannot tell you yet. No data policy has been checked against any provider's terms, so we have not read what each provider does with prompts. Prompts go to Anthropic today and sit in plain text in the workflow service's history. Checking the terms is part of the data policy, which is planned (issue #57). Use AI and stay compliant

Who is responsible when something goes wrong?

A person accepts every change, so the decision is yours. An agent cannot accept for you. Where legal responsibility falls between you and us is not written down yet, and we will not guess it here.

What happens if it fails?

A restart loses no batch work, and a failure of recurring work reaches your inbox. Whether a laptop that goes to sleep loses work has not been verified.

What will it cost?

The software is free for verified non-profits (decided 2026-10-13; a lawyer confirms the licence). You pay for onboarding, support for each active project, and yearly maintenance and support. The amounts are not decided. There is no price per user. Fit your people and your budget

We have no engineer. Is that a problem?

A separate reviewer looks at every change, and where you have no engineer, each change would say plainly that no human engineer has read it. The case that breaks the gate is accepting a change nobody on your side can judge. People and AI, building together

Do you have more ideas than developers?

Tell us. We are looking for our first NGO pilots.

Tell us you are interested

B Team is not offered to anyone yet. NGOs come first. If you want to hear when that changes, or you would like to be one of our first pilots, write to us.

Write to: [contact address to be decided]

This draft has no form and collects no data. Once there is an address, what happens next is meant to be this: we tell you whether we can work with your organisation yet. We promise no date and no price.

Next: Build what actually matters