mirror of
https://github.com/x1xhlol/system-prompts-and-models-of-ai-tools.git
synced 2026-06-17 23:09:35 +00:00
# Dealix is no longer "a platform". It is a new category: # An Autonomous Revenue Company OS that runs growth FOR Saudi businesses # as if Growth + Sales + Partnerships + Customer Success + Strategy + # Compliance + Data sat in one self-improving system. Autonomous Service Operator (16 modules) — البوت المركزي - intent_classifier: 16 supported intents (Arabic + English keywords; deterministic; no LLM) - conversation_router: route_message + handle_message — single entry point that classifies, routes to handler, recommends a bundle, builds intake + initial pipeline - session_state: 13 valid states + UUID-based sessions + audit history - intake_collector: per-intent intake question sets + parse + validation - approval_manager: Arabic approval cards (capped at 3 buttons) + decision processing (approve/edit/skip/reject including Arabic verbs) - service_orchestrator: 11-step canonical pipeline (intake→data_check→targeting→contactability→strategy→drafting→approval→execution_or_export→tracking→proof→upsell) - workflow_runner: advance + completion check - tool_action_planner: HARD-BLOCKS linkedin.scrape_profile, linkedin.auto_dm, linkedin.auto_connect, social.scrape_followers; high-risk tools require approval; draft-safe tools return draft_only; unknown tools default to approval_required - proof_pack_dispatcher: per-service Proof Pack envelope with required metrics - upsell_engine: 3 deterministic verdicts (upsell_now / iterate_first / gentle_upsell) based on csat + pipeline + meetings - whatsapp_renderer: render any card / approval / daily brief as WhatsApp draft (≤3 buttons, Arabic body, never live) - operator_memory: in-process sessions + customer_facts + preferences + audit log (production = Supabase) - service_bundles: 6 customer-facing bundles instead of 20 raw services (Growth Starter, Data to Revenue, Executive Growth OS, Partnership Growth, Local Growth OS, Full Growth Control Tower) - executive_mode: CEO command center + daily brief + revenue risks (3) + next 3 moves - client_mode: Growth Manager dashboard with 4 panels - agency_mode: multi-client roster + co-branded Proof Pack + revenue share calc Revenue Company OS (10 modules) — الذكاء عبر القنوات - event_to_card: 13 event types → Arabic decision cards (email/whatsapp/form/review/payment/risk/partner/meeting/service.completed/...) each with title_ar/summary_ar/why_now_ar/recommended_action_ar/risk_level/buttons_ar (≤3) - command_feed_engine: aggregate events for a customer + sort by risk (high first) + by_type and by_risk counts - action_graph: 14 typed edges (signal_created_opportunity → message_triggered_reply → reply_led_to_meeting → meeting_led_to_proposal → proposal_led_to_payment → ...) with what_works_for_customer scoring (outcome edges weigh more) - revenue_work_units: 19 RWU types (Salesforce-inspired): opportunity_created, draft_created, approval_collected, meeting_drafted, payment_received, risk_blocked, etc. + aggregate_work_units (counts/revenue/risks) - channel_health: cross-channel reputation snapshot (email/whatsapp/linkedin) + overall_score + channels_at_risk - opportunity_factory: turn (sector, city) into 5 opportunity cards via targeting_os.recommend_accounts + buying committee - service_factory: instantiate any service for a customer (intake + workflow + quote) - proof_ledger (revenue-tier, NOT platform_services.proof_ledger): customer-facing scoreboard with totals + summary_ar + by_type breakdown - growth_memory: anonymized cross-customer aggregates — sector_message_winrate, sector_channel_winrate, common_objections, blocked_action_reasons, successful_playbooks; best_message_for_sector + best_channel_for_sector - self_improvement_loop: weekly Arabic recommendations from real metrics (approval_rate, reply_rate, meeting_rate, blocked_actions, service_revenue) + best_service_id + next_experiment Routers (2 new) — 47 endpoints - /api/v1/operator/* (28): chat (message/decision/classify), sessions (new/transition/context/get), cards (approval/whatsapp/render), intake (questions/validate), service (start), tools (plan), proof-pack (dispatch), upsell (recommend/card), bundles (list/recommend), modes (ceo/ceo-daily-brief/ceo-risks/client/agency/agency-add-client/agency-revenue-share/agency-co-branded-proof), demos (whatsapp-daily-brief/proof-pack) - /api/v1/revenue-os/* (19): command-feed (demo/build/events-ingest), work-units (types/build/aggregate/demo), proof-ledger/demo, action-graph (edge-types/demo), channel-health (snapshot/demo), opportunity-factory (run/demo), service-factory (instantiate/demo), growth-memory/demo, self-improvement (weekly-report/demo) Tests (2 new files, 81 tests) - test_autonomous_service_operator.py: 50 tests * 8 intent classification tests (want_more_customers, has_contact_list, partnerships, whatsapp, pricing, approve, unknown fallback) * 4 conversation router (recommends correct service per intent + bundle, processes approval decisions) * 4 session lifecycle (UUID, transition validation, memory store, context build) * 4 intake (questions per intent, validation detects missing fields, complete intake passes) * 4 approval (≤3 buttons, approve/skip Arabic, unknown decision returns error) * 5 tool planner (linkedin scrape blocked, auto_dm blocked, high-risk → approval, draft-safe → draft_only, unknown → approval_required) * 4 bundles (6 total, agency → partnership_growth, local → local_growth_os, default → growth_starter) * 7 modes (CEO Arabic, daily brief 3 decisions, 3 risks, client panels, agency aggregation, revenue share calc, co-branded includes both names) * 3 WhatsApp renderer (no live send, ≤3 buttons, Arabic morning text) * 4 proof + upsell (Proof Pack draft, upsell_now for strong, iterate_first for weak, ≤3 buttons) - test_revenue_company_os.py: 31 tests * 4 event → card (email Arabic, low review high-risk, risk.blocked high, unknown → action_required) * 3 command feed (demo 8 events, sorts high-risk first, empty handling) * 4 RWUs (≥18 types, build validates, aggregate sums revenue, risks_blocked counted) * 4 action graph (≥12 edge types, validates type, demo 2 customers, what_works scoring) * 2 channel health (returns score, flags risky channel) * 2 opportunity factory (5 opps no live send, blocks unsafe in notes) * 3 service factory (instantiate known + unknown errors, demo 4 services) * 3 proof ledger (appends, rejects unknown, demo has revenue + risks) * 2 growth memory (top objections, best message per sector) * 3 self-improvement (low approval recommends fix, high blocked recommends review, returns best service) Docs (1 new + 1 updated) - AUTONOMOUS_REVENUE_COMPANY_OS.md (Arabic): 12-layer architecture + service bundles + safety + endpoints + competitive positioning - DEALIX_100_PERCENT_LAUNCH_PLAN.md: added §44 Autonomous Revenue Company OS Test results - 81/81 new tests pass - Full suite: 905 passed, 2 skipped (missing API keys, unrelated) - 0 existing tests broken Safety + integration - All 47 new endpoints: live_send_allowed=False, approval_required=True - LinkedIn scrape/auto-DM/auto-connect HARD-BLOCKED in tool_action_planner - High-risk tools (whatsapp.send_message, gmail.send, calendar.insert_event, moyasar.charge, gbp.publish_review_reply, social.publish_dm, social.publish_post) → approval_required forced - Cold WhatsApp blocked via existing contactability_matrix - Operator memory hashes nothing yet — production must wire to security_curator.trace_redactor before any persistence - 6 bundles unify the 12 productized services from Service Tower - Modes integrate platform_services + intelligence_layer + service_excellence - Action Graph + Revenue Work Units + Proof Ledger together form Dealix's Saudi Revenue Graph - Self-improvement loop reads metrics that flow from agent_observability + growth_curator Integration with everything before - Autonomous Service Operator orchestrates Service Tower, Service Excellence OS, Targeting OS, Platform Services, Intelligence Layer - Revenue Company OS reads from platform_services event_bus + intelligence_layer mission_engine + targeting_os reputation_guard - Service factory uses service_tower.get_service + build_intake_questions + quote_service - Opportunity factory uses targeting_os.recommend_accounts + map_buying_committee - Channel health uses targeting_os.calculate_channel_reputation - Tool planner integrates with platform_services.tool_gateway policies - WhatsApp renderer aligns with launch_ops button caps - Bundles map to service_tower upgrade_paths Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
8.2 KiB
8.2 KiB
Dealix Autonomous Revenue Company OS
الفئة الجديدة: Dealix ليس CRM ولا أداة واتساب ولا AI agent ولا lead scraper. هو شركة نمو رقمية ذاتية التشغيل تدخل أي بزنس، تفهمه، تبني خطة نمو، تشغّل الخدمات المناسبة، تطلب موافقات، تنسق القنوات، تفتح شراكات، ترتب اجتماعات، تجهز مدفوعات، وتثبت العائد.
1. القيم الأساسية للنظام
Signal → Context → Service Recommendation → Workflow →
Risk Check → Draft → Approval → Execution/Export →
Outcome → Proof → Learning → Upgrade
كل event داخل Dealix يمر بهذه السلسلة. لا توجد فجوة بين "إشارة" و"إيراد".
2. الطبقات الـ12
| الطبقة | الموقع |
|---|---|
| Autonomous Service Operator | auto_client_acquisition/autonomous_service_operator/ |
| Service Tower | auto_client_acquisition/service_tower/ |
| Service Excellence OS | auto_client_acquisition/service_excellence/ |
| Targeting OS | auto_client_acquisition/targeting_os/ |
| Safe Tool Gateway | auto_client_acquisition/platform_services/tool_gateway.py |
| Agent Runtime | كل layer يحدد الـ agents فيه |
| Workflow Engine | service_orchestrator + workflow_runner |
| Revenue Graph | revenue_company_os/action_graph.py |
| Proof Ledger | revenue_company_os/proof_ledger.py + platform_services/proof_ledger.py |
| Self-Improving Layer | revenue_company_os/self_improvement_loop.py + growth_curator/ |
| Revenue Launch System | revenue_launch/ + launch_ops/ |
| Growth Memory | revenue_company_os/growth_memory.py |
3. Autonomous Service Operator
16 module + 28 endpoint. البوت المركزي:
intent_classifier— 16 intent عبر Arabic + English keywords (deterministic).conversation_router— كل intent → handler + خدمة موصى بها.session_state— 13 حالة جلسة + audit history.intake_collector— أسئلة intake لكل intent + validation.approval_manager— كروت ≤3 أزرار + decisions (approve/edit/skip/reject).service_orchestrator— pipeline 11-step canonical.workflow_runner— advance + completion check.tool_action_planner— يحظر LinkedIn scraping/auto-DM، يطلب approval لـ high-risk، draft فقط للآمنة.proof_pack_dispatcher— Proof Pack envelope per service.upsell_engine— 3 verdicts (upsell_now / iterate_first / gentle_upsell).whatsapp_renderer— ≤3 buttons، Arabic body.operator_memory— sessions + facts + preferences + audit.service_bundles— 6 bundles (Growth Starter, Data to Revenue, Executive Growth OS, Partnership Growth, Local Growth OS, Full Growth Control Tower).executive_mode— CEO command center.client_mode— Growth Manager dashboard.agency_mode— multi-client + co-branded Proof Pack + revenue share.
4. Revenue Company OS
10 module + 19 endpoint. الذكاء عبر القنوات:
event_to_card— 13 event types → Arabic decision cards (≤3 buttons).command_feed_engine— daily aggregation + sort by risk.action_graph— 14 typed edges signal → action → outcome → proof.revenue_work_units— 19 RWU types (Salesforce-inspired) + aggregation.channel_health— cross-channel reputation snapshot.opportunity_factory— turn signals into opportunity cards.service_factory— instantiate any service for a customer.proof_ledger— Revenue Proof scoreboard per customer.growth_memory— cross-customer aggregates (anonymized): best message/channel/objections.self_improvement_loop— weekly Arabic recommendations from real metrics.
5. Service Bundles (6 customer-facing offerings)
| Bundle | Best for | Price (SAR) |
|---|---|---|
| Growth Starter | أي شركة تجرب لأول مرة | 499–1,500 |
| Data to Revenue | شركات لديها قائمة | 1,500–3,000 |
| Executive Growth OS | CEO / Growth Manager شهرياً | 2,999 |
| Partnership Growth | شركات تنمو عبر الشركاء | 3,000–7,500 |
| Local Growth OS | عيادات/متاجر/فروع | 999–2,999 |
| Full Growth Control Tower | مؤسسات 30+ يوم | 12,000–25,000 |
6. الأمان (Critical Gates)
كل tool action يمر:
Intent → Policy → Approval → Execution → Audit
أوضاع التنفيذ:
suggest_onlydraft_onlyapproval_requiredapproved_execute(env flag مفعّل + اعتماد)blocked
الممنوع تماماً (حتى مع env flag):
- LinkedIn scraping / auto-DM / auto-connect.
- cold WhatsApp بدون opt-in.
- Moyasar live charge من API.
- إرسال Gmail بدون اعتماد بشري.
7. Endpoints الجديدة
Autonomous Service Operator (28)
POST /api/v1/operator/chat/{message, decision, classify}
POST /api/v1/operator/sessions/{new, {id}/transition, {id}/context}
GET /api/v1/operator/sessions/{id}
POST /api/v1/operator/cards/{approval, whatsapp/render}
GET /api/v1/operator/intake/questions/{intent}
POST /api/v1/operator/intake/validate
POST /api/v1/operator/service/start
POST /api/v1/operator/tools/plan
POST /api/v1/operator/proof-pack/dispatch
POST /api/v1/operator/upsell/{recommend, card}
GET /api/v1/operator/bundles
POST /api/v1/operator/bundles/recommend
POST /api/v1/operator/mode/{ceo, ceo/daily-brief, ceo/risks, client, agency, agency/add-client, agency/revenue-share, agency/co-branded-proof}
GET /api/v1/operator/whatsapp/daily-brief/demo
GET /api/v1/operator/proof-pack/demo
Revenue Company OS (19)
GET /api/v1/revenue-os/command-feed/demo
POST /api/v1/revenue-os/{events/ingest, command-feed/build}
GET /api/v1/revenue-os/work-units/{types, demo}
POST /api/v1/revenue-os/work-units/{build, aggregate}
GET /api/v1/revenue-os/proof-ledger/demo
GET /api/v1/revenue-os/action-graph/{edge-types, demo}
POST /api/v1/revenue-os/channel-health/snapshot
GET /api/v1/revenue-os/channel-health/demo
POST /api/v1/revenue-os/opportunity-factory
GET /api/v1/revenue-os/opportunity-factory/demo
POST /api/v1/revenue-os/service-factory
GET /api/v1/revenue-os/service-factory/demo
GET /api/v1/revenue-os/growth-memory/demo
POST /api/v1/revenue-os/self-improvement/weekly-report
GET /api/v1/revenue-os/self-improvement/demo
8. اختبارات
tests/unit/test_autonomous_service_operator.py — 50 tests.
tests/unit/test_revenue_company_os.py — 31 tests.
تغطية:
- Intent classification (8 intents).
- Bundle recommendation per persona.
- Tool planner blocks LinkedIn scrape/auto-DM.
- Approval cards ≤3 buttons.
- Sessions transition + audit.
- Modes (CEO / Client / Agency) with revenue share calc.
- Event → card with risk levels.
- Action Graph what-works.
- RWU aggregation + revenue total.
- Self-improvement recommendations.
9. الفرق الشاسع عن المنافسين
| المنافس | ماذا يملك | أين Dealix يتفوق |
|---|---|---|
| CRM | بيانات وفرص | يقول ماذا تفعل اليوم |
| WhatsApp tool | إرسال | يقرر هل ترسل، لمن، ولماذا، وبأي موافقة |
| Email assistant | يكتب رد | يحول الإيميل إلى pipeline + meeting + Proof |
| Agency | تنفيذ يدوي | نظام قابل للتكرار + Proof Pack |
| Generic AI agent | ينفذ prompts | عنده خدمات + سياسات + Proof + موافقات + تحسين ذاتي |
| HubSpot/Gong/Salesforce | منصات قوية | سعودي/عربي/SMB/Service-first/WhatsApp-aware |
10. الخلاصة
Dealix الآن فئة جديدة:
- 12 طبقة معمارية متكاملة.
- 905 اختبار ناجح.
- 47 endpoint جديد في هذه الجولة.
- Approval-first في كل قناة.
- Self-improving أسبوعياً.
- Revenue Work Units قابلة للقياس.
- Proof Ledger يُثبت العائد.
- 6 bundles + Service Excellence Score يحكم كل خدمة.
لا يبيع features. يبيع نتائج منظمة.