The AI chat assistant scoped tool queries by company but ignored the
per-user Bouncer abilities the rest of the app enforces, so any `use ai`
holder could read customers, invoices, payments, and company financials
their role couldn't otherwise see. Each AiTool now declares a required
ability (entity-aligned); the registry hides unauthorized tools from the
model and refuses to execute them as a backstop.
Separately, admin/owner-supplied URLs were fetched server-side with no
guard against private/reserved targets (SSRF): the AI base URL, the
CurrencyConverter "DEDICATED" exchange-rate URL, and S3/Spaces file-disk
endpoints. A shared PrivateNetworkGuard now backs a PublicHttpUrl
validation rule (save-time) and runtime guards in each driver.
- AiTool::requiredAbility() + mapping across all 12 tools
- AiToolRegistry filters schemas() by ability and re-checks in execute()
- PrivateNetworkGuard / BlockedUrlException / PublicHttpUrl rule (new)
- Rule wired into AI config (service + 3 controllers), exchange-rate,
and file-disk endpoints; runtime guards in OpenRouterDriver,
CurrencyConverterDriver, and FileDiskService
- Tests for ability filtering, the guard, the rule, and 422 rejections
Extracts the two inline LLM prompts (AiAssistantService's chat system
prompt and AiTextGenerationService's writing preamble) out of PHP
heredocs and into plain-markdown template files under resources/ai/
prompts/. Each file can now be edited without opening a service
class, without wrestling with PHP string interpolation, and with
proper markdown syntax highlighting in editors.
A tiny PromptLoader helper at app/Support/Ai/PromptLoader.php reads
the file and does {{placeholder}} substitution via strtr() — no
Blade, because Blade's {{ $var }} HTML-escapes ampersands and quotes,
which is wrong for LLM prompts (a company called "Smith & Co" would
be sent as "Smith & Co"). Missing templates throw RuntimeException
so they fail loud during development.
Pure refactor: no prompt wording changes. Existing AI feature tests
(AiChatFlowTest, AiGenerationTest) pass unchanged — they assert on
message structure via ScriptedAiDriver, not on prompt content. Three
new unit tests in PromptLoaderTest lock the helper's contract:
placeholder substitution, no-var loading, missing-file error.
Foundation for the AI chatbot + text generation feature. Phase 1 is infrastructure only: driver plumbing, configuration storage with encrypted API keys, global vs per-company resolution, admin + company UI pages, and an optional installer wizard step. The chat assistant and text-generation WYSIWYG integration come in later phases.
**Driver plumbing (app/Support/Ai/)** — AiDriver abstract, AiDriverFactory, AiException, AiChatResponse DTO, OpenRouterDriver concrete implementation. OpenRouter is the OpenAI-compatible aggregator that unlocks hundreds of models behind one API key and one request shape — ideal as the default v1 driver. Drivers are extensible the same way exchange rate drivers are: the module Registry's generic registerDriver('ai', ...) machinery plus a typed Registry::registerAiDriver() convenience wrapper (shipped in the upstream invoiceshelf/modules package in a paired commit).
**AiConfigurationService** — mirrors MailConfigurationService shape but with one deliberate deviation: API keys are encrypted at the service layer via Crypt::encryptString before persistence. OpenRouter bearer tokens have much bigger blast radius than SMTP passwords. Same settings / company_settings tables, same global-vs-per-company pattern, same use_custom_ai_config override toggle. Resolution order: global ai_enabled must be YES, then the company either overrides via use_custom_ai_config=YES (and can opt out with ai_enabled=NO inside the override) or inherits the global config.
**Controllers** — Admin/Settings/AiConfigurationController (global CRUD + driver list + test connection), Company/Settings/CompanyAiConfigurationController (per-company override + test), Setup/AiConfigurationController (installer wizard step, skippable with explicit ai_enabled=NO). API key is always masked as '********' in GET responses — the frontend submits the placeholder back on save and the backend preserves the stored value.
**Installer wizard** — new optional step 7 'AI' between Mail and Account. Default OFF with a Skip button. MailView.vue now routes to installation.ai instead of installation.account; installation.ai then routes to installation.account. Step order comment updated in routes.ts.
**Admin + Company settings pages** — AdminAiConfigView (no toggle, always global) and AiConfigView (with use_custom_ai_config BaseSwitchSection that auto-saves OFF). Both share AiConfigurationForm which renders the driver selector, API key input with show/hide, driver-specific config_fields (base_url for OpenRouter), and per-role enable toggles with free-text model inputs backed by a datalist of suggested models from driver metadata.
**Bootstrap endpoint** — adds an ai block to the response: { enabled, chat_enabled, text_generation_enabled }. All three are booleans resolved through AiConfigurationService::resolveForCompany(). Never leaks the API key. Frontend feature flags read from bootstrapData.ai to decide whether to show Phase 2/3 UI.
**Bouncer ability** 'manage ai config' added to SettingsPolicy, gated on isSuperAdmin() (same pattern as manage email config, manage pdf config).
**Tests** (22 new) — Unit: AiDriverFactory resolves built-in + Registry-contributed drivers, rejects unknown, merges availableDrivers. AiConfigurationService: encryption round-trip, resolution order (3 cases: global off, inherit global, override with company key, override with opt-out), makeDriver null/instance cases, listDrivers metadata. Feature: admin save + read with api key masking, preserve-on-placeholder behavior, company toggle ON/OFF semantics, bootstrap ai flags reflect resolution, company opt-out path.
372 tests pass (was 350, +22). Pint clean. npm run build clean. Phase 2 (chat assistant + tool calling) and Phase 3 (WYSIWYG text generation popup) are separate follow-up commits — this one is the foundation only.
