Your first agent

Working in two weeks

Two weeks is a delivery commitment, not a sales figure.

It starts when five named conditions are met, it produces something reviewable rather than demonstrable, and production timing is specific to your proposal.

Before the clockReadiness
Five conditions, each with a concrete test you can apply to yourself. The clock starts when they are met, not when the contract is signed.
Weeks 1–2A reviewable agent
An end-to-end Agent running in Test, exercising the real Surface, the real identity, the real Connection, and the real evidence path.
After reviewEvaluation and approval
The version is qualified against your cases and thresholds, and the security, provider, and customer-approval gates are worked through.
Then continuouslyControlled launch
A deliberate promotion through a controlled rollout, monitoring that runs whether or not anyone is watching, and rollback as a release.

What the two weeks produce

A reviewable end-to-end Agent in Test, exercising the whole path rather than a slice.

The real path

A request enters through the intended Surface. The runtime resolves the Tenant, User, Agent Release, context, model, and authorized Tools.

The real execution

Structured-data work runs as a bounded one-shot program in a fresh sandbox — the same way it will in production.

The real evidence

Runs, Calculations, source references, and Evaluation results. That is what makes it reviewable rather than demonstrable.

What does not count

A prototype that bypasses the Surface, the identity, the Connection, or the evidence path proves nothing about the path production will use.

The five conditions that start the clock

Conditions rather than paperwork. Each exists because the launch fails without it, and each has a test you can apply before a proposal is signed.

One workflow and an acceptance boundary

One trigger, one actor, one outcome, explicit non-goals. Not “answer questions about our data”, and not a feature list.

Ready when Two people from different teams describe it the same way, and someone can say what an unacceptable answer looks like.

The access it will actually use

A test account or approved representative source, authenticating with the exact identity and scope the workflow will use.

Ready when Someone on your team has run the read, with that account, and can show the result.

Examples and expected results

Real cases covering normal work, hard work, missing data, and cases that should be refused — plus the vocabulary and rules behind them.

Ready when A domain reviewer can look at two candidate answers and say which is correct, and why.

The tenancy and Integration decisions

Which Organization, Project, Environment, Tenant, and User the work belongs to, and which systems it reaches.

Ready when The resource, identity, and Integration map is settled rather than assumed.

An empowered contact

A named person available across both weeks to answer questions, make decisions, and attend the review.

Ready when The person is named, and their availability is agreed rather than hoped for.

Rescoping is far cheaper before the clock than after it.

After the two weeks

Production timing

It follows Evaluation, security, provider, and customer-approval gates — and it is specific to the proposal rather than to a calendar.

Who authors the configuration

Your engineers can own Agent Configuration through the Management API, its typed client, and the first-party Cantora Terraform provider. Cantora-managed is the same platform, with Cantora authoring the initial agent.

Same API, same Runtime Plane. The route changes who authors it, and nothing else. See how Cantora computes an answer, or what each route costs.

Get started

Scope your first agent

Bring one bounded workflow, the systems it needs, and examples of a good answer. Cantora replies with the technically useful next step.

Scope your first agent