Deeply integrated. Operationally optional. Developer controlled.

Safire AI is being designed as a first-class platform service that can understand the real project, use explicit tools and produce reviewable development changes — while Safire remains fully usable with AI disabled.

Not a bolt-on chatbot

AI should work with the application model.

Safire AI can receive structured context from the selected procedure, control, window, table, query, test or project instead of relying only on pasted text.

Language and runtime knowledge generated from Safire metadata.
Project-aware source, symbol and application context.
Document and image understanding where enabled.
Typed tool calls rather than unrestricted shell access.
Review, approval, build and test evidence around changes.

Developer capabilities

Explain, design, develop, review, test and diagnose.

EXP

Explain

Understand unfamiliar source, windows, data relationships and application behaviour.

DES

Design

Help design windows, classes, data structures, APIs, reports and workflows.

DEV

Develop

Generate or modify Safire source and structured application objects.

REV

Review

Find risks, missing validation, inconsistent behaviour and change impact.

TEST

Test

Generate tests, run approved suites and analyse failures.

DOC

Document

Create and maintain technical, design and support documentation.

Provider independence

Local, remote or enterprise models behind a Safire-owned service contract.

Safire should not scatter provider-specific calls throughout the IDE or runtime.

LocalLM Studio, Ollama or OpenAI-compatible local services.

RemoteApproved commercial APIs where policy permits.

PrivateCompany-hosted or air-gapped inference.

DisabledSafire continues to build, run and maintain applications normally.

SafireIDE

Safire apps

Safire AI serviceContext · Tools · Policy · Audit

Local model

Remote provider

Permission and approval

Advice and action are different capabilities.

AI can be granted explicit levels of access. High-impact actions should remain approval-gated.

Capability	Typical policy
Read selected project context	Allowed when AI is enabled for the project.
Explain / review / plan	Allowed.
Propose source changes	Allowed, presented as a reviewable change.
Apply writes	Approval policy controls whether the AI may commit the change to working source.
Build and run tests	Allowed according to project/tool policy.
Create checkpoint / restore	Explicit developer approval.

AI for Safire applications

The same AI core can serve developers and end-user applications.

Safire applications can eventually use AI controls and language APIs without coupling their business logic to a particular model provider.

Review current AI document

Illustrative Safire AI surface

Assistant AIControl
    ProviderProfile = "LocalDevelopment"
    Context = CurrentProject
    RequireWriteApproval = True
    ShowCitations = True
End

AI works for the developer.

Safire remains a professional development environment whether AI is enabled or not.

Follow AI development