Instant Deployment Library
The Shiva Agents library provides a collection of production-ready AI personas and structural blueprints designed for immediate business use. You can deploy specialized agents across various departments or implement complex, multi-agent workflows with a single click.
Building AI agents from scratch is time-consuming and prone to errors in logic or reliability. Instead of starting with a blank screen, builders can tap into a curated ecosystem designed to handle real-world business operations. Whether you are a developer building a new SaaS or a business owner automating your back office, the library offers a head start.
- Departmental Personas: Access a library of production personas spanning over twenty departments. These are not just generic chatbots; they are specialized entities with specific roles and goals.
- Department-in-a-Box: Deploy nine "department-in-a-box" blueprints for immediate utility. These blueprints provide a pre-configured team of agents that work together to complete complex departmental tasks.
- Runnable Workflows: Utilize 23 runnable workflow patterns that are installable in one click. These patterns allow you to move from a conceptual idea to a running agentic process instantly.
For example, a small retail business in Mumbai could use a "Customer Support Department-in-a-box" to handle inquiries, manage returns, and update inventory without manual intervention for every single message.
Proven Multi-Agent Topologies
When one agent is not enough, you need a system where multiple agents collaborate. Shiva Agents provides access to proven "topologies"—the structural designs that dictate how agents communicate, divide labor, and reach a goal. These patterns are based on research and practical application in workflow patterns.
You can use the "One Team" node within the visual canvas to run any of these topologies. This node acts as a container that manages the complex interaction logic so you don't have to code it manually.
Commonly Used Topologies
Depending on your business goal, you can choose from several established communication structures:
- Orchestrator-workers: An orchestrator breaks a large task into smaller subtasks, hands them to specialized workers, and then combines the results. This is ideal for complex projects like code generation or large-scale research.
- Handoff networks: Agents pass control to the right specialist as the task evolves. Think of this like a customer service line where a generalist agent transfers your call to a technical expert.
- Assembly-line SOP: Each role hands a typed, structured artifact to the next agent in the chain, much like a manufacturing line.
- Group chat: A speaker selector picks which agent should speak next, allowing for fluid, conversational collaboration.
- Debate and judge: Multiple agents argue different sides of an issue, and a dedicated judge agent decides the final outcome. This is excellent for high-stakes decision-making or quality auditing.
- Manager and crew: A manager agent plans the work and assigns tasks to a crew of specialists.
- Role-play pairs: Two agents with fixed, opposing, or complementary roles iterate back and forth to refine a result.
- Task and progress ledgers: An orchestrator tracks facts, the current plan, and overall progress to ensure nothing is missed.
Specialized Agent Personas
In Shiva Agents, an agent is more than just a prompt. Every persona in the library is built with a specific identity to ensure consistent behavior. This is a core part of agentic architecture, where the model's ability to follow instructions is wrapped in a structured framework.
When you create or select a persona in the Agent Studio, you define several key attributes:
- Role and Goal: A clear definition of what the agent is supposed to do and what "done" looks like.
- Backstory and Style: The tone and personality the agent uses when communicating.
- Skills: Specialized capabilities such as research, design, coding, data analysis, translation, or planning.
- Tools: A specific set of functions the agent is allowed to call, such as checking a database or sending a message.
Crucially, every persona is configured with an autonomy level. This ranges from Observe (the agent only watches and reports) to Assist (the agent suggests actions) to Act (the agent performs tasks) and finally Autonomous (the agent operates with full independence within its bounds).
Visual Workflow Integration
For complex business processes, you need more than just individual agents; you need a map of how they interact. Shiva Agents provides a visual canvas, similar to tools like n8n, where you can design entire operational flows.
On this canvas, any node can represent a single agent or an entire team of agents working together. This allows you to visualize the "brain" of your business operations. The system includes 20 different node types to give you granular control over the process:
| Node Category | Examples of Node Types |
|---|---|
| Triggers | Manual, Schedule (Cron), Webhook, Event-based |
| Logic & Flow | Branch by rules, Branch by model, Fan-out map, Merge, Sub-workflow |
| Agentic Action | Agent node, Team node, Model call, Verifier loop |
| Human-in-the-loop | Approval, Ask a human, Wait for response |
| Data & System | HTTP request, Isolated code, Memory, Set state, Notify, Assign to inbox |
If building a workflow on a canvas feels too technical, you can simply describe the workflow in plain language. The architect tool will then draft the visual structure for you, significantly lowering the barrier to entry for non-technical business owners.
Reliability for Business
Most AI demos work once in a notebook but fail when applied to real business work. Shiva Agents is built for the "second world"—the world of long-running, mission-critical operations that must survive restarts and handle errors gracefully.
To ensure your business can rely on these agents, the platform implements several production-grade safeguards:
- Durability and Checkpoints: The engine creates a checkpoint after every single step. If a server restarts or a connection drops, the agent can resume exactly where it left off.
- Cost and Safety Controls: You can set token-based execution limits, bounded cycles (to prevent infinite loops), and step budgets to ensure an agent doesn't spend more than your allocated budget on a single task.
- Governance and Auditing: Every action taken by an agent is recorded in a hash-chained audit log. This creates an unchangeable record of what happened, making it impossible to quietly edit the history.
- Risk-Based Permissions: Tools are assigned risk classes (such as read, write, or critical). An agent's autonomy level must match the risk level of the tool it is trying to use.
Open Integration Standards
Shiva Agents is designed to be part of your existing tech stack, not a walled garden. It uses open protocols to ensure that your agents can talk to other software and other agents seamlessly.
How to Connect
There are three primary ways to integrate Shiva Agents into your ecosystem:
- OpenAI-Compatible API: You can call any agent or workflow using the standard
/v1/chat/completionsendpoint. This means any software that works with OpenAI can work with your Shiva Agents. You can learn more about these capabilities in the Shiva Agents API documentation. - Model Context Protocol (MCP): By using the MCP server, you can make your agents and workflows available as tools to any MCP-compliant client.
- Agent-to-Agent (A2A) Delegation: Through A2A agent cards and JSON-RPC endpoints, your agents can discover and delegate tasks to other agents, enabling a truly interconnected workforce.
By following these standards, you ensure that your AI workforce is future-proof and ready to collaborate in a growing ecosystem of intelligent software.

Questions people ask
What is included in the Shiva Agents library?
The library contains production personas for over twenty departments, nine department-in-a-box blueprints, and 23 runnable workflow patterns.
Can I use my own models with these agents?
Yes, Shiva Agents provides an OpenAI-compatible API, allowing you to call agents or workflows using any standard OpenAI client.
How do I deploy a workflow from the library?
The library patterns are designed to be installable in one click, allowing you to move from a blueprint to a running agentic process immediately.
What kind of autonomy do the agents have?
Agents have configurable autonomy levels ranging from Observe and Assist to Act and fully Autonomous, with risk-based guardrails for critical actions.
How does the library handle complex multi-agent tasks?
The library provides proven topologies like Orchestrator-workers and Handoff networks to manage how agents communicate and divide tasks.
