agentic-org

# AgenticOrg

**AI Virtual Employee Platform** — LangGraph agents, 1000+ integrations (via Composio), native connectors and tools. Agents call real APIs (Jira, HubSpot, GitHub, GSTN, Tally, Banking AA) — not just generate text. Voice agents, RAG knowledge base, smart LLM routing, industry packs, PII redaction, browser RPA, CFO/CMO dashboards, ABM dashboard, NL Query (Cmd+K), multi-company support, scheduled reports, A/B testing, email drip engine, web push HITL, Python/TypeScript SDKs, MCP server, human-in-the-loop governance, no-code builder.

**Live counts** for agents, connectors, tools, and product version come from `GET /api/v1/product-facts`. Anywhere a count appears in the UI it is fetched from that endpoint — do not hand-edit numbers in docs or pages; they will drift.

[![Live](https://img.shields.io/badge/Live-agenticorg.ai-blue)](https://agenticorg.ai)
[![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](LICENSE)
[![Python](https://img.shields.io/badge/Python-3.12-green.svg)](https://python.org)
[![React](https://img.shields.io/badge/React-18-blue.svg)](https://react.dev)
[![E2E](https://img.shields.io/badge/E2E-Playwright_against_production-brightgreen.svg)](ui/e2e/)
[![PyPI](https://img.shields.io/badge/PyPI-agenticorg-blue.svg)](https://pypi.org/project/agenticorg/)
[![npm](https://img.shields.io/badge/npm-agenticorg--sdk-blue.svg)](https://www.npmjs.com/package/agenticorg-sdk)
[![Version](https://img.shields.io/badge/Version-4.8.0-green.svg)](CHANGELOG.md)

**Live**: https://agenticorg.ai | **App**: https://app.agenticorg.ai | **Playground**: https://agenticorg.ai/playground

---

## Latest Mainline Status (2026-06-13)

- **OACP cache foundation**: C6X4 added the durable `oacp_artifact_cache_records` repository for public-safe artifact refs scoped by buyer agent, seller agent, tenant, and merchant. C6X5 added a deterministic maintenance planner over those records.
- **Commerce boundary**: OACP cache work is internal and fail-closed. It does not enable public OACP publication, live checkout, live payments, live provider rails, merchant private APIs, or production commerce readiness.
- **Security hardening**: production dependencies were trimmed, JWT runtime moved to `PyJWT[crypto]`, `python-jose`/`ecdsa` are blocked by regression gates, Docker runtime no longer needs `curl`, and security CI fails closed on high-risk findings.
- **Deployment hardening**: production rollout uses the manual Cloud Run helper with split Cloud Run and Artifact Registry regions, image digest and commit metadata checks, migration-first rollout support, and explicit traffic modes.
- **SDK launch coverage**: Python SDK, TypeScript SDK, and MCP server now cover A2A/MCP discovery, `commerce_sales_agent` launch, connector listing, knowledge search, agent generation, workflow generation, workflow creation, workflow runs, and run-status polling in regression tests.

---

## Commerce Sales Agent

AgenticOrg includes a Commerce Sales Agent that talks to Grantex Commerce V1 through Grantex-only tools. It is designed for consent-first product discovery, cart drafting, Commerce Passport handling, provider-neutral payment intent handoff, checkout handoff, and payment status polling without calling payment providers directly.

> Commerce production readiness is not final. AgenticOrg production MCP/A2A commerce metadata may be visible, but Grantex production Commerce V1 discovery remains disabled/fail-closed and live checkout, live payments, and live Plural are not enabled.

```mermaid
flowchart LR
  user[User question] --> agent[Commerce Sales Agent]
  agent --> tools[grantex_commerce:* tools]
  tools --> grantex[Grantex Commerce REST/MCP]
  grantex --> catalog[Catalog and inventory]
  grantex --> consent[Consent and Commerce Passport]
  grantex --> payment[Provider-neutral payment control]
  payment --> mock[Mock provider smoke evidence]
  payment -. blocked .-> live[Live provider / Plural gate]
```

| Area | Current posture |
| --- | --- |
| Default runtime mode | Mock mode remains the default for local demo use. |
| Real-staging mode | Explicitly gated and only valid against approved Grantex staging or exact temporary smoke URLs. |
| Hosted discovery smoke | API-only C3 smoke verified liveness, health, MCP tools, and A2A discovery. |
| Provider boundary | No direct Stripe, Plural, Pine, or provider credential commerce path. |
| Production discovery caveat | AgenticOrg commerce metadata should stay gated or explicitly reviewed until Grantex read-only discovery is approved. |
| OACP cache foundation | C6X1-C6X5 cover verifier planning, cache runtime, repository boundary, durable SQL-backed cache records, and maintenance planning. Internal only. |
| Durable cache storage | C6X4 stores public-safe source/evidence refs with TTL, freshness, revocation snapshot, risk tier, non-enablement flags, tenant-safe indexes, and RLS. |
| Cache maintenance planner | C6X5 classifies records into keep, refresh, evict, purge, quarantine, source refresh, or human-review outcomes. Planner only; no scheduler or side effects. |
| Live checkout/payments | Blocked; do not imply production payment readiness. |

Read more in `docs/commerce-agent-overview.md`, `docs/commerce-agent-developer-guide.md`, `docs/commerce-agent-hosted-staging-e2e.md`, `docs/reports/commerce-agent-c6x4-durable-oacp-cache-repository.md`, `docs/reports/commerce-agent-c6x5-oacp-cache-maintenance.md`, `docs/reports/commerce-agent-real-staging-evidence.md`, `docs/reports/commerce-agent-hosted-smoke-evidence.md`, and `docs/reports/commerce-agent-production-discovery-readiness.md`.

Key tool aliases: `grantex_commerce:merchant_get_profile`, `grantex_commerce:catalog_search`, `grantex_commerce:catalog_get_item`, `grantex_commerce:inventory_check`, `grantex_commerce:cart_create`, `grantex_commerce:consent_request`, `grantex_commerce:consent_exchange`, `grantex_commerce:payment_create_intent`, `grantex_commerce:checkout_create`, and `grantex_commerce:payment_get_status`.

## What Is AgenticOrg?

AgenticOrg deploys **AI virtual employees** that automate enterprise back-office work. Each agent has a name, designation, specialization, and tailored instructions — like a real team member. They process invoices, run payroll, triage support tickets, launch campaigns, and reconcile bank statements — with human approval on every critical decision.

### Key Numbers

| Metric | Value |
|--------|-------|
| Pre-built Agents | See `GET /api/v1/product-facts` `agent_count` — live from registry |
| Custom Agents | Unlimited via no-code wizard |
| Enterprise Connectors | **1000+ integrations** (via Composio) + native connectors (`connector_count` on `/product-facts`), `tool_count` total tools — all with real API endpoints |
| Dashboards | CFO Dashboard + CMO Dashboard (role-specific KPI views) |
| NL Query | Cmd+K search bar + slide-out chat panel with agent attribution |
| Multi-Company | Company switcher for CA firms managing multiple client entities |
| Scheduled Reports | Celery beat --> PDF/Excel --> email/Slack/WhatsApp delivery |
| Workflow Templates | Production-ready templates in `core/workflows/templates` |
| Prompt Templates | Managed via `/api/v1/prompt-templates` |
| Automated Tests | Backend pytest + frontend vitest + Playwright E2E against production (no skipped tests) |
| CI E2E | Enabled against production on every push to `main` |
| SDKs | Python (`pip install agenticorg`), TypeScript (`npm i agenticorg-sdk`), MCP Server, CLI |
| LLM | Smart routing via RouteLLM: Gemini Flash (free) / Gemini Pro / Claude/GPT-4o. Air-gapped: Ollama/vLLM |
| Deployment | Cloud Run manual helper (`scripts/deploy_cloud_run.sh`), Cloud Run in `asia-southeast1`, Artifact Registry in `asia-south1` by default |
| Version | Sourced from `pyproject.toml` / `GET /api/v1/product-facts` |

### What It Does

| Before | After |
|--------|-------|
| 5-day month-end close | **1.5 days** with AP + Recon + Close + Treasury agents |
| Manual invoice processing | **11 seconds** per digital PDF invoice (parse --> GSTIN --> 3-way match --> GL) |
| 40% ticket mis-routing | **88% auto-classification** accuracy |
| 2-week employee onboarding | **4 hours** (Darwinbox + Slack + email auto-provisioned) |
| Manual bank reconciliation | **99.7% auto-match** rate, done by 6 AM daily |
| Zero payroll errors | PF, ESI, TDS computed automatically |
| CFO drowning in reports | **CFO Dashboard**: Cash Runway, DSO, DPO, AR/AP Aging at a glance |
| CMO guessing ROI | **CMO Dashboard**: CAC, ROAS by Channel, MQL/SQL pipeline in real time |
| "Ask the analyst" queries | **NL Query**: Cmd+K --> "What's my cash position?" --> instant answer |
| CA firm juggling 20 clients | **Multi-company switcher**: one login, all client entities |
| Manual report emails | **Scheduled Reports**: Celery --> PDF/Excel --> email/Slack/WhatsApp |

### Agents That Act, Not Just Talk

Unlike AI chatbots that only generate text, AgenticOrg agents **execute real actions**:

```
You: "Production API returning 500 errors. CloudSQL connection pool exhausted."

Ops Commander (Aria Singh):
  → Gemini reasons about severity (636 tokens, 2.6s)
  → Creates Jira ticket KAN-5 via real API (1.1s)
  → Sets priority to Highest, assigns engineering team
  → Returns analysis + ticket link to you
```

Verified on production: agents have created **14 real Jira tickets**, read **HubSpot CRM data** (contacts, deals, companies), and queried **GitHub repo statistics** — all through the LLM → tool_calls → connector → API pipeline.

---

## Architecture

```
Landing Page (animations, interactive demo, blog, SEO, developer section)
    |
App Dashboard
    ├── CFO Dashboard (/dashboard/cfo) — Cash Runway, Burn Rate, DSO, DPO, AR/AP Aging
    ├── CMO Dashboard (/dashboard/cmo) — CAC, MQLs, SQLs, Pipeline, ROAS by Channel
    ├── NL Query (Cmd+K search bar + slide-out chat panel)
    ├── Multi-Company Switcher (CA firms managing multiple clients)
    ├── Agent Fleet, Workflows, Appr