What is an Agent Library?
The Agent Library within the Shiva Agents platform provides a collection of pre-configured AI personas and proven multi-agent workflow patterns. It is designed to help businesses move from concept to production quickly by offering specialized agents for various departments.
Think of it as a starting point for your digital workforce. Instead of spending weeks trying to prompt an AI to act like a specialized accountant or a customer support lead, you can access these roles immediately. The library includes over twenty departmental personas and nine "department-in-a-box" blueprints, allowing you to deploy entire functional units of AI agents to handle complex business processes.
- A collection of pre-configured AI personas and runnable workflow patterns.
- Includes over twenty departmental personas and nine "department-in-a-box" blueprints.
- Designed for one-click installation to move from concept to production quickly.
Proven Multi-Agent Patterns
When one agent is not enough to complete a complex task, you need a way for multiple agents to work together. In a production environment, you don't just want a chatbot; you want a system where agents follow a structured logic to achieve a goal. Shiva Agents provides 23 runnable workflow patterns to manage these interactions.
For example, a small business in Mumbai managing logistics might use an "Assembly-line SOP" pattern, where one agent processes incoming orders, a second agent checks warehouse stock, and a third agent drafts the shipping labels. Each agent hands a "typed artefact" (a specific, structured piece of data) to the next one in the sequence.
| Pattern Name | How it Works |
|---|---|
| Assembly-line SOP | Each role hands a typed artefact to the next agent in sequence. |
| Group chat | A speaker selector picks the next agent to speak. |
| Debate and judge | Agents argue over a solution while a third agent decides the outcome. |
| Manager and crew | A manager agent plans the work and assigns tasks to specialists. |
| Role-play pair | Two agents with fixed roles iterate back and forth to solve a problem. |
| Handoff network | Agents pass control to the right specialist as the situation requires. |
| Task and progress ledgers | An orchestrator tracks facts, plans, and overall progress. |
These patterns are part of a larger agentic architecture, which ensures that the model is wrapped in a loop of planning, acting, observing, and remembering.
Department-in-a-Box Blueprints
For many business owners, managing individual agents can feel overwhelming. This is why the library offers "Department-in-a-Box" blueprints. These are specialized, high-level templates that cover entire functional areas of a business.
Instead of building a single agent to handle one task, these blueprints allow you to deploy entire functional units of AI agents instantly. You can choose from nine specialized blueprints that cover more than twenty different departmental personas. This approach is ideal for businesses looking to automate entire workflows, such as a complete "Customer Success Department" or a "Sales Operations Unit," rather than just individual tasks.
By using these blueprints, you are essentially hiring a digital team that already knows how to communicate and coordinate, rather than just hiring a single freelancer.
From Library to Durable Workflows
The true power of the Agent Library is realized when these agents are integrated into durable, visual workflows. Shiva Agents uses an n8n-style canvas where you can draw out your business processes. In this environment, any node on the canvas can be a single agent or an entire team of agents working from the library.
This allows for a hybrid approach: you can use a standard workflow automation AI to handle predictable, code-based steps, and then drop in an agent from the library when you need "reasoning" or flexibility. For instance, a workflow could start with a webhook trigger (a customer sends an email), move through a branch by rules, and then pass the task to a "Manager and Crew" team from the library to resolve a complex complaint.
You can combine library agents with several different node types to ensure reliability:
- Triggers: Start a process via a schedule, a webhook, or a manual command.
- Verifiers: Use a loop to check if an agent's work meets specific criteria before proceeding.
- Human Approval: Pause the workflow to ask a human for permission before performing a risky action, like issuing a refund.
- Sub-workflows: Nest complex agentic processes within larger business operations.
This structure is what separates a simple AI agent platform from a production-grade system. It ensures that if a server restarts or a process takes two days to get a human's approval, the agent's memory and progress are preserved.
Open Protocols and Connectivity
The Agent Library is not a walled garden. It is designed to be part of a larger, interconnected ecosystem. To ensure that your agents can talk to other software and other AI systems, Shiva Agents utilizes several open standards.
First, all agents are accessible via an OpenAI-compatible API. This means any existing software designed to work with OpenAI can interact with your library agents by simply changing the endpoint. If you have a custom application, you can call an agent as if it were a standard model.
Second, the library supports the Model Context Protocol (MCP). This allows agents and workflows to be available as tools to any MCP client, making it easy for your agents to discover and use new tools without massive reconfiguration.
Finally, the platform uses A2A (Agent2Agent) cards. An A2A card, located at a standard endpoint, allows other agents to discover your library agents and delegate tasks to them. This creates a world where different AI systems can "hire" each other to get a job done.
Production-Grade Governance
As businesses move from experimenting with AI to relying on it for daily operations, safety and oversight become critical. You cannot have an agent making unauthorized payments or leaking sensitive data. Shiva Agents implements governance through several layers of control.
Every single action taken by an agent is recorded in a hash-chained audit log. This creates a permanent, uneditable record of exactly what the agent did, making it possible to review its decisions long after the task is finished. This is essential for compliance and troubleshooting.
To manage risk, the platform uses Autonomy Levels. Every tool an agent can use is assigned a risk class: read, fetch, write, external, or critical. You can set rules that dictate how much freedom an agent has based on these classes. For example, an agent might have "Observe" autonomy for reading files, but require "Human Approval" before it can use a "critical" tool like sending a payment. This allows you to move from "Assist" to "Autonomous" operations safely.
| Feature | Benefit for Business Owners |
|---|---|
| Hash-chained Audit Log | Provides a tamper-proof history of all agent actions for total accountability. |
| Autonomy Gating | Limits tool access based on risk, preventing unauthorized or expensive actions. |
| Data Residency Guard | Ensures sensitive data (like Indian personal data) stays within required geographic boundaries. |
| Spending Limits | Prevents runaway costs by setting strict token and step budgets on every run. |
Frequently Asked Questions
- Can I customize the agents in the library?
- Yes, you can use the library as a starting point and customize the role, goal, backstory, and tools in the Agent Studio.
- How do I deploy an agent from the library?
- The library is designed for one-click installation, making it easy to add specialized agents to your organization.
- Do these agents work with existing software?
- Yes, agents can use tools via MCP, A2A, or OpenAI-compatible APIs to interact with other systems.
- How are the multi-agent patterns structured?
- The library includes patterns like 'Manager and crew' or 'Handoff networks' to manage how agents communicate and delegate tasks.
- Is there a way to monitor agent actions?
- Yes, every action is captured in a hash-chained audit log to ensure a permanent, uneditable record of what the agent did.

Questions people ask
Can I customize the agents in the library?
Yes, you can use the library as a starting point and customize the role, goal, backstory, and tools in the Agent Studio.
How do I deploy an agent from the library?
The library is designed for one-click installation, making it easy to add specialized agents to your organization.
Do these agents work with existing software?
Yes, agents can use tools via MCP, A2A, or OpenAI-compatible APIs to interact with other systems.
How are the multi-agent patterns structured?
The library includes patterns like 'Manager and crew' or 'Handoff networks' to manage how agents communicate and delegate tasks.
Is there a way to monitor agent actions?
Yes, every action is captured in a hash-chained audit log to ensure a permanent, uneditable record of what the agent did.
