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

---

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	GKE Autopilot, ~$95/month
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 invoice (OCR --> 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, Approvals, Audit, Sales Pipeline, Integrations
    └── Report Scheduler (create, manage, toggle, run-now scheduled reports)
    |
FastAPI Backend
    ├── Agent Registry (36 agents) → Smart LLM Router (RouteLLM)
    │       ↓ tool_calls                    ↑ synthesis
    │   Tool Gateway → 1000+ Integrations (Composio) + Native Connectors (see /api/v1/product-facts for counts)
    │       ├── Jira (11 tools) ← verified, creates real tickets
    │       ├── HubSpot (13 tools) ← verified, reads real CRM
    │       ├── GitHub (9 tools) ← verified, reads real repos
    │       ├── GSTN (8 tools) ← real Adaequare 2-step auth + DSC signing
    │       ├── Tally (7 tools) ← XML/TDL protocol + bridge for on-premise
    │       ├── Banking AA (5 tools) ← RBI-compliant consent flow
    │       ├── Bombora, G2, TrustRadius ← intent data (weighted 40/30/30)
    │       └── GA4, MoEngage, NetSuite, WordPress, Twitter/X, YouTube, Mailchimp...
    ├── NL Query Engine → Agent attribution → Chat history
    ├── Scheduled Report Engine → Celery beat → PDF/Excel → email/Slack/WhatsApp
    ├── LangGraph Runtime → GraphInterrupt HITL → Shadow Mode
    ├── Workflow Engine (20+ templates) → real agent execution → HITL Queue
    ├── NEXUS Orchestrator → Audit Logger
    ├── A2A Protocol → Agent Discovery → Cross-platform Tasks
    ├── RAGFlow Engine → Document ingestion → Semantic search → Agent retrieval
    ├── LiveKit + Pipecat → Voice agents → STT/TTS → SIP telephony
    ├── RouteLLM → Smart model routing → 3-tier cost optimization
    ├── Presidio → Pre-LLM PII redaction → Aadhaar/PAN/GSTIN scrubbing
    ├── Composio → 1000+ tool marketplace → OAuth bridge
    ├── MCP Server → native tools exposed to Claude/Cursor/ChatGPT (see /api/v1/product-facts.tool_count)
    ├── API Key Manager → ao_sk_ keys → SDK/CLI/MCP auth
    ├── SOP Parser → Upload SOPs → Deploy as agents
    └── Sales Agent → Gmail API → Email Sequences
    |
SDKs: Python (PyPI) | TypeScript (npm) | MCP Server | CLI
    |
PostgreSQL (Cloud SQL) + Redis + GCS + GCP Secret Manager

Agent Domains (6 Domains, 50+ Agents)

Domain	Agents	Key Agents
Finance	10	AP Processor, AR Collections, Reconciliation, Tax Compliance, Month-End Close, FP&A, Treasury, Expense Manager, Rev Rec (ASC 606), Fixed Assets
HR	6	Talent Acquisition, Onboarding, Payroll Engine, Performance Coach, L&D, Offboarding
Marketing	9	Content Factory, Campaign Pilot, SEO Strategist, CRM Intelligence, Brand Monitor, Email Marketing, Social Media, ABM, Competitive Intel
Operations	5	Support Triage, Contract Intelligence, Compliance Guard, IT Operations, Vendor Manager
Back Office	3	Risk Sentinel, Legal Ops, Facilities Agent
Comms	3	Ops Commander (Jira triage), DevOps Scout (GitHub + Jira health), Slack Notifier
Sales	1	Automated sales agent (qualification, email sequences, pipeline)
Industry Packs	16	Healthcare (4), Legal (4), Insurance (4), Manufacturing (4) — one-click install
Custom	37+ on demo	Create via 5-step no-code wizard — unlimited

Pre-built agents across 6 domains + 4 industry packs. 1000+ integrations (via Composio) + native connectors with tools. All endpoints real. Live counts: GET /api/v1/product-facts.

---

Core Features

CFO Dashboard (/dashboard/cfo)

Role-specific KPI view for finance leaders. Real-time metrics powered by finance agents:

• Cash Runway — months of cash remaining at current burn rate
• Burn Rate — monthly cash outflow trend
• DSO / DPO — Days Sales Outstanding and Days Payable Outstanding
• AR/AP Aging — receivables and payables bucketed by 30/60/90/120+ days
• P&L Summary — revenue, COGS, gross margin, EBITDA
• Bank Balances — aggregated from all connected bank accounts via AA
• Tax Calendar — upcoming GST/TDS/advance tax deadlines with status

CMO Dashboard (/dashboard/cmo)

Role-specific KPI view for marketing leaders. Real-time metrics powered by marketing agents:

• CAC — Customer Acquisition Cost by channel
• MQLs / SQLs — Marketing and Sales Qualified Leads pipeline
• Pipeline Value — total opportunity value by stage
• ROAS by Channel — Return on Ad Spend for Google, Meta, LinkedIn
• Email Performance — open rate, CTR, unsubscribe, deliverability
• Brand Sentiment — positive/negative/neutral trend from social monitoring
• Content Performance — top pages by traffic, engagement, conversions

NL Query Interface

Natural language search across all your business data:

• Cmd+K Search Bar — global shortcut opens search from any page
• Slide-out Chat Panel — full conversational interface with agent attribution
• Agent Attribution — every answer shows which agent(s) provided the data
• Example Queries: "What's my cash position?", "Show me AP aging over 90 days", "How did Google Ads perform last week?", "What's our DSO this quarter?"

Multi-Company Support

Built for CA firms and holding companies managing multiple entities:

• Company Switcher — dropdown in the top nav to switch between client entities
• Isolated Data — each company has its own agents, connectors, workflows, and audit trail
• Cross-Company Reporting — consolidated views across all managed entities
• RBAC per Company — different roles and permissions for each entity

Scheduled Report Engine

Automated report generation and delivery:

• Celery Beat Scheduler — cron-based scheduling for any report
• Output Formats — PDF and Excel with branded templates
• Delivery Channels — email, Slack, and WhatsApp
• Report Scheduler UI — create, manage, toggle on/off, and run-now from the dashboard
• Pre-built Schedules — daily cash report, weekly P&L, monthly close package, weekly marketing digest

Workflow Templates (15 Pre-built)

Production-ready workflow templates that combine multiple agents:

Template	Domain	Description
invoice_to_pay_v3	Finance	OCR --> GSTIN --> 3-way match --> payment execution
month_end_close	Finance	Trial balance --> adjustments --> reconciliation --> close
daily_treasury	Finance	Cash position --> sweep --> forecast --> report
tax_calendar	Finance	Deadline tracking --> filing prep --> DSC signing
campaign_launch	Marketing	Brief --> content --> review --> publish --> monitor
content_pipeline	Marketing	Ideation --> draft --> SEO --> approval --> publish
lead_nurture	Marketing	Scoring --> segmentation --> drip --> wait_for_event --> handoff to sales
weekly_marketing_report	Marketing	Collect metrics --> build report --> deliver
email_drip_sequence	Marketing	Behavior-triggered email sequences (open/click/time delays)
ab_test_campaign	Marketing	Create variants --> run test --> auto-winner selection --> CMO override --> send winner
abm_campaign	Marketing	CSV upload targets --> intent scoring (Bombora/G2/TrustRadius) --> personalized outreach
incident_response	Ops	Triage --> Jira ticket --> assign --> monitor --> resolve
lead_to_revenue	Sales	Qualify --> outreach --> follow-up --> close
weekly_devops_health	Ops	GitHub + Jira metrics --> health score --> report

Virtual Employee System

Each agent is a virtual employee with persona, specialization, and routing:

• Employee name (Priya, Arjun, Maya) — appears in audit logs and approvals
• Designation (Senior AP Analyst - Mumbai) — role clarity
• Specialization (Domestic invoices < 5L) — task routing
• Multiple agents per role — 3 AP Processors for different regions with smart routing

Organization Chart

Visual tree hierarchy per department. Each agent has an org_level field (e.g., Head → Manager → Analyst) and a parent reference, forming a reporting chain used for smart escalation — when an agent's confidence is below threshold, the task auto-escalates to its parent in the org tree. CSV bulk import lets you onboard entire departments in one upload.

No-Code Agent Creator

5-step wizard: Persona → Role → Prompt → Behavior → Deploy as Shadow. Or describe the employee you need in plain English — the system auto-generates the full agent configuration.

NL Workflow Builder

Describe business processes in plain English (e.g., "Automate invoice approval when amount > 5L") and the system generates the complete workflow definition. Preview before deploying.

Multi-Language Support

Platform available in English and Hindi. Language picker in header. Agents respond in the user's preferred language. Extensible to Tamil, Telugu, Kannada.

Content Safety

AI-generated content is checked for PII leakage, toxicity (via HuggingFace toxic-bert), and near-duplicate detection before delivery. Configurable thresholds per agent.

Prompt Template Library

27 production-tested templates with {{variable}} substitution. Built-in templates are read-only — clone to customize. Full prompt audit trail with edit history. Prompt lock on active agents.

Human-in-the-Loop (HITL)

Configurable confidence floors, trigger conditions, escalation chains, and timeout rules. Every HITL decision logged in audit trail. Agents cannot bypass their own gates. LangGraph-based HITL uses GraphInterrupt for pause/resume — the agent graph pauses at the approval node and resumes only after human decision.

RBAC

• CEO/Admin: Full access, all domains
• CFO: Finance agents only
• CHRO: HR agents only
• CMO: Marketing agents only
• COO: Operations agents only
• Auditor: Read-only audit log

Sales Agent

Automated lead qualification, personalized email outreach, follow-up sequences (Day 1/3/7/14), Gmail inbox monitoring, and pipeline dashboard. Demo request form → instant personalized response.

Native Enterprise Connectors — All Real API Endpoints

Finance (11): Oracle Fusion, SAP, Tally (XML/TDL + bridge), GSTN (Adaequare 2-step + DSC), Stripe, QuickBooks, Zoho Books, Banking AA (RBI consent), Income Tax India, Pine Labs (Plural), NetSuite HR (8): Darwinbox, Okta, Greenhouse, LinkedIn Talent, DocuSign, Keka, Zoom, EPFO Marketing (19): Salesforce, HubSpot, Google Ads, LinkedIn Ads, Meta Ads, Ahrefs, Mixpanel, Buffer, Brandwatch, GA4, MoEngage, WordPress, Twitter/X, YouTube, Mailchimp, Semrush, Bombora, G2, TrustRadius Ops (7): Jira, ServiceNow, Zendesk, PagerDuty, Confluence, Sanctions API, MCA Portal Comms (9): Slack, GitHub, Gmail, SendGrid, GCS, Google Calendar, Twilio, WhatsApp, LangSmith

Every connector uses real API endpoints from official documentation. Zero stubs. The Tally connector uses native XML/TDL protocol with a bridge agent for on-premise instances. Banking AA follows RBI-compliant consent flow. GSTN uses Adaequare's 2-step authentication with DSC signing for filing.

Tier 1: Marketing Automation (v3.2.0)

Feature	What It Does
Web Push Notifications	One-tap approve/reject HITL decisions from browser push notifications — no need to open the dashboard
Email Drip Engine	Behavior-triggered email sequences with open/click/time-delay steps and automatic re-engagement for non-openers
A/B Testing	Create campaign variants, auto-select winners by open rate or CTR, with CMO override before sending to remaining audience
Email Webhooks	Real-time SendGrid, Mailchimp, and MoEngage open/click tracking via inbound webhooks
Intent Data	Bombora + G2 + TrustRadius aggregation with weighted scoring (40/30/30) for account-level buying signals
ABM Dashboard	Target account management with intent heatmap, CSV upload, tier filtering, and one-click campaign launch
Wait Step	Real time delays in workflows (was stub) — supports minutes, hours, and day-based delays
Wait-for-Event	Pause workflow execution until a specific event occurs (email opened, link clicked, form submitted)

CA Firms -- Paid Add-on

INR 4,999/month per client | 14-day free trial

Purpose-built for Chartered Accountant firms managing multiple client companies from a single tenant.

Feature	Details
5 AI Agents	GST Filing, TDS Compliance, Bank Reconciliation, FP&A Analyst, AR Collections
Partner Dashboard	Aggregate KPIs across all clients -- filing status, overdue counts, client health scores
Filing Approvals	Agent generates filing --> pending --> partner self-approve --> filed. Supports bulk approve across clients
GSTN Credential Vault	AES-256 encrypted storage per client. Passwords never returned in API responses. Supports key rotation via encryption_key_ref
Compliance Calendar	Auto-generated deadlines for GSTR-1, GSTR-3B, TDS 26Q/24Q. Email alerts at 7-day and 1-day before due date
Tally Auto-Detect	POST /api/v1/companies/tally-detect reads Tally bridge and auto-creates the company entity with GSTIN, PAN, and FY
Demo Seeded	7 demo companies seeded for demo@cafirm.agenticorg.ai -- ready to explore immediately

---

Quick Start

Prerequisites

• Python 3.11+ (tested on 3.12, 3.13) | Node.js 20+ | Docker

Installation

AgenticOrg provides three requirements files for different deployment scenarios:

File	Packages	Purpose	Required?
requirements.txt	45 pinned	Core platform (API, agents, workflows, connectors, auth)	Yes
requirements-v4.txt	4	v4 features: Composio 1000+ tools, RouteLLM smart routing, Presidio PII redaction	Optional (graceful degradation)
requirements-dev.txt	11	Development: pytest, ruff, mypy, bandit, pre-commit	For contributors only

All versions in requirements.txt are pinned to exact production-tested versions (2026-04-06).

Local Development

git clone https://github.com/mishrasanjeev/agentic-org.git
cd agentic-org
cp .env.example .env  # Add your Gemini API key (free at aistudio.google.com)

# Install Python dependencies
pip install -r requirements.txt              # Core platform (required)
pip install -r requirements-v4.txt           # v4 features (optional)
pip install -r requirements-dev.txt          # Dev tools (optional)

# Start infrastructure (PostgreSQL + Redis)
docker compose up -d postgres redis

# Run API server
uvicorn api.main:app --host 0.0.0.0 --port 8000 --reload

# Run frontend (separate terminal)
cd ui && npm install && npm run dev

Optional Docker Services

docker compose --profile ragflow up -d       # Knowledge Base (RAGFlow)
docker compose --profile voice up -d         # Voice Agents (LiveKit)
docker compose --profile airgap up -d        # Local LLM (Ollama, CPU)
docker compose --profile airgap-gpu up -d    # Local LLM (vLLM, GPU)

Docker Compose

docker compose up -d
# API: http://localhost:8000 | UI: http://localhost:3000

GKE Production

gcloud container clusters get-credentials agenticorg-lean --region=asia-south1
helm upgrade --install agenticorg ./helm -n agenticorg -f helm/values-lean.yaml

Demo Credentials

Role	Email	Password
CEO	ceo@agenticorg.local	ceo123!
CFO	cfo@agenticorg.local	cfo123!
CHRO	chro@agenticorg.local	chro123!
CMO	cmo@agenticorg.local	cmo123!
COO	coo@agenticorg.local	coo123!
Auditor	auditor@agenticorg.local	audit123!

---

API Reference

Base URL: https://app.agenticorg.ai/api/v1

Method	Path	Auth	Description
GET	/health/liveness	No	Health check
POST	/auth/signup	No	Create account
POST	/auth/login	No	Get JWT token
POST	/auth/google	No	Google OAuth
POST	/auth/forgot-password	No	Request password reset email
POST	/auth/reset-password	No	Reset password with token
POST	/demo-request	No	Demo form → lead + sales agent
GET	/agents	JWT	List agents (RBAC filtered)
POST	/agents	JWT	Create agent (tools auto-populated by type/domain)
POST	/agents/{id}/run	JWT	Execute agent
PATCH	/agents/{id}	JWT	Update (prompt lock on active)
GET	/agents/org-tree	JWT	Org chart tree (department hierarchy)
POST	/agents/import-csv	JWT	Bulk import agents via CSV
POST	/agents/{id}/clone	JWT	Clone with persona
POST	/agents/{id}/promote	JWT	Shadow → Active (shadow limit enforcement)
GET	/agents/{id}/prompt-history	JWT	Prompt audit trail
GET	/prompt-templates	JWT	List templates
POST	/prompt-templates	JWT	Create template
GET	/sales/pipeline	JWT	Sales pipeline
POST	/sales/pipeline/process-lead	JWT	Trigger sales agent
POST	/sales/process-inbox	JWT	Process Gmail replies
GET	/sales/metrics	JWT	Weekly digest data
GET	/workflows	JWT	List workflows
GET	/approvals	JWT	HITL approval queue
GET	/audit	JWT	Audit log
GET	/kpis/cfo	JWT	CFO dashboard KPIs
GET	/kpis/cmo	JWT	CMO dashboard KPIs
POST	/chat/query	JWT	NL Query (natural language question)
GET	/chat/history	JWT	Chat history for current user
GET	/companies	JWT	List companies (multi-company)
POST	/companies	JWT	Create company entity
PATCH	/companies/{id}	JWT	Update company entity
GET	/report-schedules	JWT	List scheduled reports
POST	/report-schedules	JWT	Create scheduled report
PATCH	/report-schedules/{id}	JWT	Update scheduled report
POST	/report-schedules/{id}/run-now	JWT	Trigger immediate report run
DELETE	/report-schedules/{id}	JWT	Delete scheduled report
GET	/connectors	JWT	List native connectors (count from /product-facts)
GET	/connectors/registry	JWT	Connector registry (all registered connectors + tool counts)
GET	/connectors/{id}/health	JWT	Connector health check
GET	/connectors/{id}	JWT	Connector details
PUT	/connectors/{id}	JWT	Update connector config
POST	/org/api-keys	Admin	Generate API key (ao_sk_ prefix)
GET	/org/api-keys	Admin	List API keys
DELETE	/org/api-keys/{id}	Admin	Revoke API key
GET	/a2a/agent-card	No	A2A agent discovery card
POST	/a2a/tasks	JWT/Grantex	Execute A2A task
GET	/mcp/tools	No	List MCP tools (see /api/v1/product-facts.tool_count)
POST	/mcp/call	JWT/Grantex	Call MCP tool
POST	/sop/upload	JWT	Upload and parse SOP document
POST	/sop/parse-text	JWT	Parse SOP text
POST	/sop/deploy	JWT	Deploy parsed SOP as agent
POST	/agents/{id}/delegate	JWT	Grantex delegation setup

SDKs, CLI & MCP Server

Python SDK

pip install agenticorg

from agenticorg import AgenticOrg

client = AgenticOrg(api_key="ao_sk_your_key_here")
result = client.agents.run("ap_processor", inputs={"invoice_id": "INV-001"})
agents = client.agents.list()
sop = client.sop.parse_text("When invoice > 5L, require CFO approval")
card = client.a2a.agent_card()

TypeScript SDK

npm install agenticorg-sdk

import { AgenticOrg } from "agenticorg-sdk";

const client = new AgenticOrg({ apiKey: "ao_sk_your_key_here" });
const result = await client.agents.run("ap_processor", { inputs: { invoice_id: "INV-001" } });
const agents = await client.agents.list();

MCP Server

Any MCP-compatible client (Claude Desktop, Cursor, ChatGPT) can use AgenticOrg agents and tools:

AGENTICORG_API_KEY=ao_sk_... npx agenticorg-mcp-server

Or add to your MCP client config:

{
  "mcpServers": {
    "agenticorg": {
      "command": "npx",
      "args": ["agenticorg-mcp-server"],
      "env": { "AGENTICORG_API_KEY": "ao_sk_..." }
    }
  }
}

CLI

pip install agenticorg

agenticorg agents list
agenticorg agents run ap_processor --input '{"invoice_id": "INV-001"}'
agenticorg sop parse "When invoice > 5L, require CFO approval"
agenticorg mcp tools

See sdk/README.md and mcp-server/README.md for full documentation.

---

API Keys

API keys provide programmatic access for SDKs, CLI, and MCP integrations.

• Prefix: All keys use the ao_sk_ prefix (e.g., ao_sk_a1b2c3...)
• Generation: Admin users generate keys from Settings > API Keys in the app, or via POST /api/v1/org/api-keys
• Admin-only: API key endpoints require the agenticorg:admin scope
• Security: Keys are bcrypt-hashed at rest; the full key is shown only once at creation
• Revocation: Keys can be revoked instantly from Settings or via DELETE /api/v1/org/api-keys/{id}

---

Integration Protocols

A2A (Agent-to-Agent)

AgenticOrg implements Google's A2A protocol for cross-platform agent discovery and task execution:

• GET /a2a/agent-card — public agent discovery card (no auth required)
• POST /a2a/tasks — execute tasks via A2A protocol (JWT or Grantex auth)

MCP (Model Context Protocol)

Full MCP server exposing the live connector tools to any MCP-compatible client (see /api/v1/product-facts.tool_count):

• GET /mcp/tools — list all available MCP tools (no auth required)
• POST /mcp/call — call any MCP tool (JWT or Grantex auth)

Grantex Authorization & Scope Enforcement

OAuth2-based authorization with manifest-based scope enforcement (Grantex SDK v0.3.3+):

• Manifest-based permissions: 53 pre-built tool manifests define exact permission levels (READ/WRITE/DELETE/ADMIN) for every connector tool — no keyword guessing
• grantex.enforce(): Offline JWT verification + manifest permission check in <1ms per tool call (JWKS cached after first call)
• Permission hierarchy: admin > delete > write > read — write scope covers read, admin covers everything
• LangGraph enforcement: validate_scopes graph node checks every tool call before execution
• ToolGateway enforcement: API-direct tool calls also use grantex.enforce() as primary enforcement
• POST /agents/{id}/delegate — set up Grantex delegation for agent-to-agent trust
• Scope hierarchy enforced: child agents cannot elevate parent scopes
• Token pool with automatic refresh at 50% TTL
• Custom manifests: place JSON/YAML files in GRANTEX_MANIFESTS_DIR for custom connectors
• Environment variables: GRANTEX_API_KEY (required), GRANTEX_BASE_URL (optional), GRANTEX_MANIFESTS_DIR (optional)

Environment Variables

COMPOSIO_API_KEY              — 1000+ tool integrations (optional)
AGENTICORG_LLM_ROUTING        — auto | tier1 | tier2 | tier3 | disabled
AGENTICORG_LLM_MODE           — cloud | local | auto
AGENTICORG_PII_REDACTION_MODE — before_llm | logs_only | disabled
RAGFLOW_API_URL               — RAGFlow instance URL for knowledge base
RAGFLOW_API_KEY               — RAGFlow API key
LIVEKIT_URL                   — LiveKit server URL for voice agents
LIVEKIT_API_KEY               — LiveKit API key
LIVEKIT_API_SECRET            — LiveKit API secret
STRIPE_SECRET_KEY             — Stripe billing key (hosted tier)
STRIPE_WEBHOOK_SECRET         — Stripe webhook signing secret

Knowledge Base (RAG)

Upload company documents (PDF, Word, Excel) and agents query them via semantic search. Powered by RAGFlow (Apache 2.0). Agents use knowledge_base_search tool automatically.

Voice Agents

Real-time voice AI with telephony support. Default STT: Whisper (local, Apache 2.0). Default TTS: Piper (local, MIT). Configurable SIP provider (Twilio/Vonage). Powered by LiveKit (Apache 2.0) + Pipecat (BSD-2).

Smart LLM Routing

Automatic multi-model routing via RouteLLM (Apache 2.0). Three tiers: Economy (Gemini Flash, free), Standard (Gemini Pro), Premium (Claude/GPT-4o). Reduces costs by 85% while maintaining accuracy. Air-gapped mode routes to Ollama/vLLM.

Pre-LLM PII Redaction

Microsoft Presidio (MIT) scrubs sensitive data before it reaches the LLM. Custom India recognizers: Aadhaar, PAN, GSTIN, UPI. Configurable: AGENTICORG_PII_REDACTION_MODE=before_llm|logs_only|disabled.

Industry Packs

Pre-built agent bundles: Healthcare (4 agents, HIPAA-aware), Legal (4 agents), Insurance (4 agents), Manufacturing (4 agents). One-click install via POST /packs/{name}/install.

Browser RPA

Playwright-based automation for legacy web portals without APIs. Pre-built scripts for Indian government portals (EPFO, MCA, Income Tax). Sandboxed Docker execution with screenshot audit trail.

Billing & Hosted Tier

Self-hosted is free forever. Hosted tier: Free (3 agents) / Pro ($49/mo) / Enterprise ($299/mo). India pricing: Free / Rs 999/mo / Rs 4999/mo. Stripe (global) + PineLabs Plural (India).

Explainable AI

Every agent decision includes a "Why?" panel with 3-5 plain-English bullet points, confidence bar, tools cited, and Flesch-Kincaid readability scoring. Non-technical users can understand why the agent approved or rejected an action.

Self-Improving Agents

Thumbs up/down feedback on every agent run. After 10+ feedback entries, the system auto-generates prompt amendments. Agents learn from corrections without manual prompt editing.

Multi-Agent Collaboration

Workflow collaboration step type runs 2+ agents in parallel with shared context. Aggregation strategies: merge (combine outputs), vote (majority wins), first_complete (fastest agent wins).

Customer Support Deflection

Pre-built support_deflector agent auto-resolves 60%+ of support tickets using FAQ matching + knowledge base (RAG) search. Tracks deflection rate metric on the dashboard.

Real-Time CDC (Change Data Capture)

Webhook receivers + polling-based CDC for connected systems. When data changes externally (e.g., new HubSpot deal), CDC events trigger workflows automatically. HMAC-SHA256 signature validation, fail-closed security.

Integration Workflow Page

The /integration-workflow page provides a visual guide for connecting external systems, with SDK/CLI quickstart examples and protocol documentation.

---

Testing

# Backend tests (1931+ total)
pytest tests/

# Frontend vitest (93 tests)
cd ui && npx vitest run

# Playwright E2E — 17 spec files, CI runs against production
npx playwright test

# Production connector test — real Jira/HubSpot/GitHub API calls
python tests/test_production_connectors.py

Suite	Tests	What It Covers
Backend (pytest)	See CI	Unit, security, connector harness (every native connector × exposed tools), synthetic data, regression, integration, voice, RAG, RPA, packs
Frontend (vitest)	93	Component tests, hooks, utilities
Playwright E2E	17 spec files	Full browser flows: login, onboarding, agents, workflows, approvals, landing, SOP, dashboards, ABM, drip, A/B, push
CI E2E	Enabled	Runs against production on every merge to main

---

Project Structure

agenticorg/
├── api/v1/                 # FastAPI endpoints (agents, auth, sales, templates, connectors, api-keys, sop, a2a, mcp)
├── core/
│   ├── agents/             # 50+ agent types + 27 prompt templates
│   │   ├── prompts/        # Production system prompts
│   │   ├── registry.py     # Agent registry (built-in + custom type fallback)
│   │   └── base.py         # BaseAgent: LLM reasoning → tool calling → HITL
│   ├── langgraph/          # LangGraph agent graph, runner, Grantex auth, LLM factory
│   ├── orchestrator/       # NEXUS: task routing, smart routing, state machine
│   ├── llm/                # LLM router (Gemini primary, Claude/GPT-4o fallback)
│   ├── models/             # SQLAlchemy ORM (agents, workflows, HITL, leads, templates, api_keys)
│   ├── tool_gateway/       # Scope enforcement, rate limiting, PII masking, audit
│   ├── gmail_agent.py      # Gmail API integration (inbox monitor, send replies)
│   └── email.py            # SMTP email sending
├── connectors/             # Native enterprise connectors (counts: /api/v1/product-facts)
│   ├── finance/            # Oracle, SAP, GSTN, Stripe, Tally, Banking AA, NetSuite... (11)
│   ├── hr/                 # Darwinbox, Okta, Greenhouse, EPFO, Zoom... (8)
│   ├── marketing/          # Salesforce, HubSpot, Google Ads, GA4, MoEngage, Bombora, G2, TrustRadius... (19)
│   ├── ops/                # Jira, Zendesk, ServiceNow, PagerDuty... (7)
│   ├── comms/              # Slack, GitHub, Gmail, SendGrid, GCS, Twilio... (9)
│   └── framework/          # BaseConnector, auth adapters, circuit breaker
├── auth/                   # Grantex middleware, registration
├── workflows/              # Workflow engine, triggers, conditions
├── sdk/                    # Python SDK (pip install agenticorg) + CLI
├── sdk-ts/                 # TypeScript SDK (npm i agenticorg-sdk)
├── mcp-server/             # MCP Server (npx agenticorg-mcp-server)
├── ui/src/
│   ├── pages/              # Landing, Dashboard, Agents, Sales, Blog, Resources, Integrations, IntegrationWorkflow, ConnectorCreate, ConnectorDetail, Settings
│   ├── components/         # AgentCard, ActivityTicker, InteractiveDemo, SocialProof
│   ├── pages/blog/         # 8 SEO blog articles
│   └── pages/resources/    # 26 SEO content pages across 7 topic clusters
├── tests/
│   ├── unit/               # Unit tests
│   ├── security/           # Security tests
│   ├── connector_harness/  # Connector tests (every native connector × exposed tools)
│   ├── regression/         # Regression tests
│   ├── integration/        # Integration tests
│   ├── synthetic_data/     # Invoice, resume, contract test data
│   └── e2e/                # Playwright browser tests
├── ui/e2e/                 # 17 Playwright E2E spec files
├── ui/tests/               # 93 vitest component tests
├── migrations/             # PostgreSQL DDL files
├── helm/                   # Kubernetes Helm charts
├── docs/                   # PRD, architecture, QA test plan
└── scripts/                # Seed data, deployment helpers

---

SEO & AI Discoverability

Asset	URL	Status
Landing page	https://agenticorg.ai	39 URLs in sitemap
Blog (8 articles)	/blog	Invoice processing, RPA vs AI, reconciliation, HITL, no-code, month-end close, ROI measurement, CFO story
Resources (26 pages)	/resources	7 topic clusters, FAQ schemas, internal linking
llms.txt	/llms.txt	Product summary for AI crawlers (version from pyproject)
llms-full.txt	/llms-full.txt	18.7KB complete documentation
JSON-LD	7 schemas	Organization, Product, FAQ, Breadcrumb, SoftwareCompany, WebSite, SoftwareApplication
AI crawlers	robots.txt	GPTBot, ClaudeBot, PerplexityBot, ChatGPT-User, OAI-SearchBot, cohere-ai, Applebot-Extended

---

Pricing

Self-hosted is free forever — all agents, all connectors, unlimited tasks.

Plan	Price	Agents	Integrations	Tasks
Free (self-hosted)	$0	50+	1000+	Unlimited
Free (hosted)	$0	3	20	500/day
Pro (hosted)	$49/mo (Rs 999/mo India)	Unlimited	1000+	Unlimited
Enterprise (hosted)	$299/mo (Rs 4999/mo India)	Unlimited	1000+	Unlimited + SLA

Payments: Stripe (global) + PineLabs Plural (India: NEFT/RTGS/IMPS/UPI).

---

India-First Connectors

Built for Indian enterprise — not retrofitted:

• GSTN — e-Invoice IRN, GSTR-1/2A/3B/9 filing, ITC reconciliation
• EPFO — ECR filing, UAN verification, passbook download
• Income Tax India — TDS 26Q/24Q, Form 16A, 26AS credit check
• Darwinbox — HRMS (payroll, attendance, performance, onboarding)
• Tally Prime — Native XML/TDL protocol + bridge agent for on-premise tunneling (vouchers, trial balance, GST reports)
• Banking AA — RBI-compliant Account Aggregator consent flow (read-only: statements, balances, transactions)

---

Security & Compliance

Feature	Implementation
Tenant isolation	PostgreSQL RLS, Redis key namespacing, GCS prefix isolation
HITL governance	Configurable thresholds, prompt lock on active agents, escalation chains, GraphInterrupt
Audit trail	Every agent action logged, 7-year WORM retention, HMAC signed
PII masking	Default-on masking of email, phone, Aadhaar, PAN, bank accounts
SOC-2 ready	Password policy (bcrypt 12), token blacklist, rate limiting, HSTS, CSP
Prompt audit	Every prompt edit logged with user, timestamp, before/after, reason
API key security	ao_sk_ prefix, bcrypt-hashed at rest, admin-only endpoints, instant revocation
Secret management	GCP Secret Manager integration for connector credentials (secret_ref field)
Auth failure tracking	IP-based failure tracking, auto-block on threshold, auth failure clearing on success
Shadow limit enforcement	Agents must pass shadow quality gates before promotion to active
Tool validation	Tool scope enforcement — agents cannot call tools outside their authorized scope

---

Documentation

Document	Description
PRD	Complete product requirements (7 pages)
API Reference	Full API docs with Mermaid diagrams
CFO Guide	CFO user guide — dashboard, agents, NL query, reports
CMO Guide	CMO user guide — dashboard, agents, NL query, campaigns
CA Firm Setup	End-to-end CA firm deployment guide
QA Test Plan	65 manual test cases with steps
QA Test Cases	70 automated test results
Architecture	8-layer system design
Python SDK	Python SDK + CLI documentation
MCP Server	MCP Server setup and usage
Changelog	Version history and release notes
Roadmap	Current status and future plans

---

Contributing

Fork → branch → test → PR. See CONTRIBUTING.md.

• Python: ruff check . && mypy --ignore-missing-imports .
• TypeScript: npx tsc --noEmit
• Tests: pytest tests/unit/ tests/connector_harness/

License

Apache License 2.0 — free for commercial use. See LICENSE.

---

Built by Edumatica Pvt Ltd | Bengaluru, India