The Oracles argue it. Five named humans gate it.
CARI Forge turns your requirements into an approved build spec — and never releases it without a named human approving every stage. Read how below, or tell us what you want to build on the right and see it start.
Meet The OraclesWhat do you want to build using AI?
A regulated buyer’s question is all it takes to get started — write as much detail as the problem needs. The Oracles read every submission in full before any code is written, and the Elder Oracle settles anything they can’t agree on.
A real human will reply — within 48 hours during the working week.
From proof to production
A 21-day proof and a production handoff are two different promises.
The 21-Day Forge is bounded on purpose — a named problem, a finished, ready-to-use solution, and an evidence-based decision, inside three weeks. What happens after a case clears the last gate is a separate, connected stage, not an implied extension of the same 21 days.
The 21-Day Forge
Turns one named need into a governed, working proof and a decision you can act on. Agents 1–5 run it, gated by a named human at every stage.
- · A testable problem statement, with named must-not-happen clauses
- · A readiness audit and a build-versus-buy comparison
- · A workflow design with human checkpoints and escalation paths
- · A governance spec confirming the controls hold
- · An approved Blueprint and Runbook, with an audit trail and a hash chain
Production Forge
Starts only once a case has cleared the Software Build gate. Run by the two wraparound agents — Partner and Impact — not part of the 21-day clock.
- · Partner hands the approved build to your own infrastructure, with your integrator or cloud partner, so your own people run it — not us
- · Impact reads back whether the named must-not-happen clauses held once the build was live, and whether the case's own economics showed up in production
Deeper integration work, formal security review, and ongoing operations are scoped per engagement once a case reaches this stage — they are not a default inclusion of the 21-day proof.
The Oracles
The Oracles of CARI Forge — five voices, fixed remits, opposing defaults.
The Oracles each argue from a single angle. They open objections by default, not by exception. If The Oracles cannot settle the case after one round of debate, the Elder Oracle stops the run and asks you, the human.
Risk
Risk oracle. Default stance: Surfaces compliance gaps, model risk, and breach exposure first.
“What breaks if this fails in production?”
Demand
Demand oracle. Default stance: Pressure-tests evidence for the problem against pilot data.
“Is there a real user, not just a stakeholder?”
Growth
Growth oracle. Default stance: Separates one-off asks from durable capability.
“Does solving this unlock anything downstream?”
Competition
Competition oracle. Default stance: Demands build versus buy be decided explicitly, by name.
“Could a credible off-the-shelf tool already do this?”
Money
Money oracle. Default stance: Refuses numbers that depend on the AI behaving differently later.
“Is the unit economics honest at the scale we expect?”
Build · Test first · Walk away
The Elder Oracle never overrides an unresolved objection. When The Oracles give three verdicts — build it, test it first, or walk away — and one of them is contested, the human is the tiebreaker. The Elder Oracle’s decision and its rationale are written into the case file before the forge moves on.
Core agents
Five specialist agents operate the pipeline, with Partner and Impact alongside — distinct from The Oracles.
The Oracles are the five-voice governance council that audits the brief before code. Five specialist agents are the runtime that actually operates the 21-day delivery pipeline: Discovery, Readiness, Workflow, Governance, AI Build run the five pipeline stages (stage 5 is itself called Software Build, operated by the AI Build agent). Partner and Impact work alongside delivery — Partner lands the runnable build on the buyer’s infrastructure; Impact measures the change in the world it was meant to make.
- № 01
- № 02
- № 03
- № 04
- № 05
- № 06
- № 07
How this maps back to the rest of CARI Forge. The five specialist agents, with Partner and Impact alongside, are what actually run a case. The Oracles are the five voices (Risk, Demand, Growth, Competition, Money) that read the case file and rule before the pipeline starts. The Elder Oracle is the chair that ties the rule back to a named human. Three distinct governance concepts — one delivery runtime.
Each agent’s full operational boundaries — inputs, tools, outputs, prohibited actions, and success measures — along with the five-stage gate detail and technical architecture, are visible once you’re signed in, on the pipeline detail page.
The five-stage gate breakdown, systems of record, and technical architecture live inside the platform.
What each of the five gates checks, how the audit trail and hash chain work, what CARI Forge does and doesn’t touch in your own systems, and the five-layer architecture diagram — all visible once you’re signed in, not published for anyone who visits.
Sign in to see the pipeline detailBefore you apply
What actually fits inside a 21-day proof — and what doesn't.
A bounded proof needs a bounded case. Checking fit before applying saves everyone a returned Discovery gate.
Good fit
- · A real workflow or service need with a named owner
- · Accessible sample data or representative documents to test against
- · A decision that can stay human-controlled where it needs to
- · A measurable operational or customer outcome
- · A case bounded enough to evaluate inside 21 days
Not a good fit
- · A generic chatbot with no defined workflow or named owner
- · A speculative model comparison with no business outcome attached
- · A request for uncontrolled, fully autonomous decisions
- · A use case with no lawful or approved access to the data it needs
- · A full enterprise transformation, presented as a 21-day solution
Open the file. Leave a brief.
Read the council, scroll the stages, then come back up and submit. The form is the only call to action on this page — everything else is explanation.
Need to reach a named human before submitting? cari-forge@polsia.app or read the FAQ for what a regulated buyer asks first.