Frequently asked questions

MANAV is the AI Operating Platform for building an AI society — a team of AI employees that work, meet, debate, and hand off tasks under human oversight. Every action worth stopping for goes through an approval gate, and every step leaves an evidence trail. Run AI with proof. Founded in India, built for founders and lean teams worldwide.

MANAV AI is the AI workforce platform where anyone builds a team of AI agents that plan, debate, and hold meetings together. "MANAV AI" and "MANAV" refer to the same product — MANAV is the brand, and MANAV AI is the everyday name people use when they search for us.

No — MANAV is not an AGI (artificial general intelligence). MANAV is an AI Operating Platform: the layer that lets you build, orchestrate, and govern a team of AI agents on top of today's language models. AGI describes a hypothetical system with general human-level reasoning across every domain, and no such system exists yet. MANAV is designed for the opposite posture — narrow, role-scoped AI agents that stay under human oversight, with approvals and audit trails for anything consequential. When people search "MANAV AGI" they usually mean the same product as MANAV AI.

MΛNΛV (मानव) is the Sanskrit word for "human". Named to reflect our mission: AI that amplifies humans, not replaces them. Every agent on the platform runs under human-approvable guardrails and full observability so humans stay in control.

Yes. MANAV is an Indian-founded AI Operating Platform, built for founders and lean teams worldwide. Aligned with India's stated vision for human-first, sovereign, accessible AI.

Sign up for the free tier, open the visual agent composer, name your agent, give it a system prompt describing its role, pick skills and tools it needs (Slack, Gmail, GitHub, Salesforce, etc.), optionally attach a knowledge base, and pick which LLM to use with your own API key. Deploy in minutes. No code required. Or start from a ready-made agent (Marketing, Sales, Engineering, Finance, HR, Legal, Procurement) and customize.

An AI workforce is a team of specialist AI agents that coordinate on real work. On MANAV: create your specialist agents (one per role — SDR, recruiter, marketer, ops), give each their own tools and knowledge, then orchestrate them via workflows and multi-agent primitives (delegation, debate, meetings). Humans stay in the loop for risky decisions. Start with one agent, add more as you gain confidence.

A basic agent takes 10-30 minutes: define its role, pick 3-5 tools, attach a knowledge base if needed, choose an LLM, deploy. A production-grade agent (with guardrails, RBAC, audit workflow, integrations to your live systems) typically takes a few hours to a day. Multi-agent workflows connecting several agents to real business processes often ship in a week or two.

No. MANAV is no-code from day one — visual composer for agents, workflows, and knowledge bases. Everything a founder or ops lead needs is drag-and-drop. Engineers can go deeper: build custom MCP tool servers, define workflows in YAML, use the Python or TypeScript SDK, and integrate with your own infrastructure. Both audiences supported, neither excluded.

Attach a knowledge base to your agent. Sources can include Notion, Google Drive, Dropbox, S3, Confluence, GitHub docs, Postgres tables, uploaded PDFs and CSVs, or scraped websites. MANAV ingests, embeds, indexes, and grounds the agent's answers in your data automatically via native RAG. The agent cites its sources. You control which agents have access to which knowledge bases via RBAC.

Every major integration is available as a pre-built MCP (Model Context Protocol) tool server. In your workspace tools panel, enable Slack, Gmail, or Salesforce (and a hundred others — HubSpot, GitHub, Linear, Notion, Stripe, QuickBooks, Zendesk, etc.), authenticate once, then assign the tool to any agent that needs it. The agent can then send messages, read emails, update CRM records, and so on — bounded by RBAC and rate limits you set.

In the agent composer, open the guardrails tab and pick from pre-built policies (PII redaction, spend caps, prompt-injection defense, allowed / blocked topics, tool allow-lists, approval-required actions) or write your own declarative rules. Guardrails run at multiple layers so a clever prompt can't bypass them. Every guardrail hit is logged and visible in the audit trail.

An AI Operating Platform is the runtime + tools for building, deploying, orchestrating, and governing AI agents at scale. Think of it as the operating system for your AI workforce: agents run on it, tools plug into it, humans oversee it via observability and guardrails. Similar category to Salesforce Agentforce, Microsoft Copilot Studio, and Google Vertex AI Agent Builder — but vendor-neutral and made for founders instead of Fortune 500.

An AI workforce is a coordinated team of AI agents doing real work across functions — marketing, sales, engineering, finance, HR, legal, procurement, operations. Not a single chatbot. A network of specialist agents, each with their own tools, knowledge, guardrails, and role, that collaborate the way a human team does: delegating, debating, holding meetings, escalating. Humans stay in charge of the important calls.

Multi-agent AI is when several AI agents collaborate on a task rather than a single LLM handling everything. Different agents specialize in different roles (planner, researcher, coder, reviewer, executor) and coordinate via delegation, debate, or structured meetings. Multi-agent systems outperform single-agent systems on complex work because they can parallelize, cross-check each other, and combine domain expertise. MANAV treats multi-agent as a first-class runtime primitive.

Model Context Protocol is the open standard for how AI agents call external tools, introduced by Anthropic in 2024 and adopted industry-wide in 2026 (OpenAI, Google, Microsoft, Salesforce). Before MCP, every framework had its own tool-calling format. MCP unifies them: write your tool once as an MCP server, and any MCP-native host — Claude Desktop, Cursor, MANAV — can use it. MANAV is MCP-native from day one, which means no vendor lock-in on tools.

A specialist subagent is an AI agent focused on a narrow task or domain that a higher-level agent delegates to. Instead of one giant agent trying to do everything, MANAV lets you build a supervisor agent that delegates to specialist subagents (e.g., an SDR agent delegates lead research to a research subagent, cold outreach to a writing subagent, follow-ups to a scheduling subagent). Each subagent has its own tools, prompts, and guardrails.

Human-in-the-loop means the AI agent pauses at defined moments and requires a human to approve, reject, or edit the next action before proceeding. Critical for high-stakes actions like sending customer emails, committing code to main, transferring money, or publishing content. MANAV builds HITL in as a core primitive — you decide per-action which risk level requires human sign-off, and the agent waits.

BYOLLM = Bring Your Own LLM. Instead of MANAV routing your prompts through our shared LLM account, you plug in your own OpenAI, Anthropic, Google Gemini, Mistral, or self-hosted Llama API key. Your data goes directly to your model provider. MANAV never sees or trains on your prompts. Full model portability — swap providers anytime without rebuilding your agents. See our full guide: /blog/bring-your-own-llm-byollm-guide

Agentic AI is goal-oriented software: you give it an objective, and it plans, uses tools, and takes multi-step action until the job is done. It is different from generative AI, which produces content in response to a single prompt. Agentic AI is almost always built on top of a generative model — the generative model is the brain, the agentic layer adds plans, tools, memory, and guardrails. Full breakdown: /blog/agentic-ai-vs-generative-ai-difference-2026

An AI agent is a single autonomous worker that can plan, act, and use tools to complete a defined task. An AI workforce is many agents working together as a coordinated team — with roles, hand-offs, cross-department collaboration, and shared memory. Start with one agent, prove the pattern, then grow it into a workforce. Full explainer: /blog/what-is-an-ai-workforce-complete-guide-2026

AI agent meetings are structured working sessions between multiple AI agents from different roles — for example, a finance agent, a legal agent, and a sales agent — to reach a decision together. Each meeting has an agenda, a chair, and produces documented minutes. Meetings beat single-agent thinking on cross-functional decisions because each agent challenges the others' assumptions. See how they work: /blog/ai-agent-meetings-how-multi-agent-meetings-work

In MANAV: (1) pick a role template — sales follow-up, financial analyst, legal reviewer, support triage — (2) attach only the tools it needs, granted read-only where possible, (3) enable guardrails (send-approval, spend cap, domain lock), (4) test with one real task from your work, (5) turn on a schedule. Total time: about 15 minutes for a working agent. Step-by-step: /blog/how-to-build-an-ai-agent-no-code-15-minutes

Yes. The default MANAV agent is designed as an executive-level Chief of Staff for solo founders and small teams: manages your inbox, prepares meeting briefs, coordinates other agents, tracks OKRs, escalates decisions to you, and keeps a durable memory of what's happening across your business. Delegates function-specific work (marketing content, sales outreach, hiring pipeline) to specialist agents you spin up as you grow.

MANAV can run a full multi-channel marketing function with a lean human oversight layer: content calendar planning, blog and email drafting, LinkedIn and X scheduling, ad campaign optimization, SEO tracking, funnel analytics, competitor monitoring. Founders typically start with one Marketing agent, add specialist subagents (SEO researcher, email writer, ad ops) as needs grow. Humans still own brand strategy and creative direction.

Yes — pre-built agents for common roles: AI SDR (outbound sequences, lead qualification, meeting booking), AI recruiter (sourcing, screening, interview scheduling, offer coordination), AI CFO agent (books closing, spend monitoring, forecasting, board reporting). Each ships with function-specific tools pre-wired (Apollo + Attio for SDR; Greenhouse + Ashby for recruiter; QuickBooks + Ramp + Stripe for CFO). Customize prompt and guardrails to your business.

Eight departments: Marketing, Sales, Engineering, Finance, HR, Legal, Procurement, plus an executive Chief of Staff layer that coordinates across them. Each department ships with ready-made specialist agents you can deploy immediately, or use as templates to build your own. Additional agents (customer support, product management, ops) available on the Agent Store and expandable via the visual composer.

Yes. Give the agent your help docs + product knowledge base + integrations to Zendesk / Intercom / Freshdesk, and it can triage tickets, draft responses grounded in your docs, escalate complex issues to human agents, and update your knowledge base when it spots gaps. Human approval gates for refunds, cancellations, and other high-stakes actions.

Three key differences. First, MANAV is vendor-neutral — bring your own LLM key (OpenAI, Anthropic, Google, self-hosted Llama), deploy anywhere, no CRM / Microsoft 365 / Google Cloud dependency. Second, MANAV is founder-priced — free tier and fair regional pricing worldwide, not enterprise contracts targeted at Fortune 500. Third, MANAV is multi-agent-first — collaboration primitives (agent debate, meetings, delegation) are built into the runtime rather than bolted on later.

LangChain and LangGraph are Python libraries — powerful, but require an engineer to write and maintain code. MANAV is a full platform: no-code visual composer, multi-tenant runtime, RBAC, observability, guardrails, marketplace, and pre-built departments — all production-ready. Under the hood MANAV uses LangGraph-compatible orchestration, so engineers can migrate off if they ever want. For a solo founder who doesn't want to hire a Python engineer just to run agents, MANAV is the faster path.

CrewAI and AutoGen are open-source multi-agent frameworks — you write Python, you deploy the infrastructure, you build the UI, you handle observability and RBAC yourself. MANAV provides all of that as a managed platform with a no-code composer, so lean teams ship in days not months. The multi-agent primitives (debate, meetings, delegation) MANAV ships with are inspired by the same research literature — just made production-ready and no-code.

Zapier AI and n8n AI are workflow-automation tools that added AI features — great for linear trigger-action automations. MANAV is agent-first: agents reason about which tools to call and in what order, hold debates, coordinate with other agents, and adapt to changing conditions. Lindy AI is closer to MANAV (agent-native, no-code) but single-agent-focused; MANAV's differentiator is the multi-agent workforce model and deeper governance (audit, guardrails, RBAC, BYOLLM).

ChatGPT, Claude, and Gemini are conversational AI — great for asking questions, terrible for doing production work. They can't be given persistent roles, RBAC, audit trails, or specialized tools; can't coordinate with other agents; can't be governed with guardrails or human-approval gates; and they route your data through a shared consumer product. MANAV takes the same underlying LLMs (any of them — you bring the key) and wraps them in an agent platform with persistent identity, tools, memory, coordination, and governance.

In regional Postgres databases (India, EU, US, APAC available) with per-tenant isolation at database, storage, and vector layers. If you use BYOLLM, your prompts and completions go directly to your chosen model provider (OpenAI, Anthropic, etc.) and never sit in MANAV's shared storage. Data residency is configurable — pick the region you legally must operate in. Right to erasure in one click.

No. Contractually — never. Your data (prompts, agent conversations, knowledge base contents, tool call inputs and outputs) is not used to train any model — ours or a provider's. With BYOLLM, your data flows directly to your model provider under your account. Even for evaluation, benchmarking, and product improvement, we only use aggregated non-identifying metrics — never your content.

SOC 2 Type II certification is on our compliance roadmap for 2026. GDPR compliance is designed-in from day one (data residency, right to erasure, DPA available, sub-processor list on the trust page). DPDPA (India's Digital Personal Data Protection Act) is the same story — designed-in given our Indian-founded origin. HIPAA and other vertical certifications on request for Enterprise+ deployments.

Yes — the Enterprise+ tier includes self-hosted deployment (your Kubernetes cluster, your VPC, your data never leaves your infrastructure) plus white-label options. Available for regulated industries (banking, healthcare, defense, government) and any org that requires air-gapped or sovereign-cloud deployments. Contact us via the /contact page for a scoped deployment plan.

Every agent action is captured in the tamper-proof audit trail — LLM call inputs and outputs, tool calls, delegations, human-approval decisions. When a mistake happens: (1) you see exactly what went wrong and why in the observability layer, (2) the outcome is reversible if the action was a workflow step (roll back via checkpoint), (3) you can edit the agent's system prompt or guardrails to prevent recurrence, and (4) the failure feeds the continuous-learning loop that tunes the agent over time.

Free tier starts at $0/forever with 3 agents, using your own LLM keys. Pro at $79/month for 20 agents. Team at $299/month for 50 agents including RBAC. Enterprise at $799/month for 100 agents with SSO / SAML. Enterprise+ is custom-priced with on-premise deploy and white-label options. Fair regional pricing worldwide — see /pricing for details.

Free tier — not a trial. $0 forever with up to 3 agents, unlimited runs (your LLM keys pay the model bill), full access to the visual composer, MCP tool ecosystem, and multi-agent primitives. Upgrade only when you need more agents, more workspaces, RBAC, SSO, or Enterprise features. No credit card required to start.

Fair regional pricing across every tier reflects local purchasing power for Indian users. Recognized Indian startup registrations (DPIIT-certified startups) get an additional discount on Pro and Team tiers — contact us via /contact with your DPIIT number. Students and academic institutions get free access to Pro tier with a valid .edu / academic email — apply via /contact.

Yes — no code required. Compose skills, tools, knowledge bases, safety rules, and a role visually. Give the agent a system prompt. Deploy in minutes. Or start with a ready-made agent from the C-suite (Marketing, Sales, Engineering, Finance, HR, Legal, Procurement) and customize.

The Agent Store is the marketplace where creators publish pre-built agents, workflows, and knowledge packs that anyone can install into their workspace with one click. Categories cover every function — SDR agents pre-wired for specific CRMs, recruiter agents pre-wired for specific ATS, industry-specialized agents (fintech-compliance, healthcare-intake, e-commerce ops), and more. Free and paid listings.

Yes. If you've built a specialized agent (fintech-compliance, GDPR-processing, e-commerce SEO, etc.), publish it on the Agent Store. Set your own price (one-time, subscription, or free). MANAV takes a platform fee; creators keep most of the revenue. Builder economy is Pillar 3 of the MANAV Constitution — value flows to creators, not just the platform.

Yes. Multi-agent collaboration is a core primitive, not an add-on. Agents delegate tasks to specialist subagents, hold structured meetings with agendas and takeaways, and debate different approaches before committing to one. Every debate and meeting is captured in the observability layer so you can review the reasoning behind any decision.

Through four layers. RBAC per agent — each agent has explicit permissions scoped by workspace and resource. Policy guardrails — declarative rules for what agents may and may not do, enforced at multiple layers so prompt injection can't bypass. Human-in-the-loop approval gates — risky actions pause for human sign-off before executing. Tamper-proof audit log — every action, every LLM call, every tool invocation cryptographically hashed for compliance.

Yes. Bring your OpenAI, Anthropic, Google, or self-hosted Llama key. Your data never touches a shared model provider on MANAV's behalf. Full model portability — swap providers anytime without changing your agents.

MANAV comes from the Sanskrit word मानव, which means "human." We chose it because our whole design premise is that AI should amplify humans, not replace them — every agent runs under policy guardrails, every risky action can be paused for human approval, and every decision is visible in an audit log. The name reflects that principle: humans stay in the loop.

Free tier — sign up, get your first 3 agents (or use ready-made ones from the C-suite), bring your own LLM key, deploy in minutes. See /pricing for tier details or /platform for a full walkthrough of how the MANAV AI Operating Platform works.