DocsAgents
Core Concepts

AI Agents

In CipherSense Agents, an Agent is more than just a prompt. It is a stateful, autonomous entity capable of reasoning, using tools, and making decisions within your enterprise workflows.

Anatomy of an Agent

Intelligence Layer

The core LLM (GPT-5.4, Claude 4.6, etc.) that handles reasoning and natural language processing.

Toolsets (MCP)

Capabilities granted to the agent via the Model Context Protocol to interact with external data and APIs.

Guardrails

System-level constraints that define the agent's persona, safety limits, and operational boundaries.

Multi-Model Agnosticism

CipherSense allows you to assign specific models to specific tasks. Use high-reasoning models like Claude Sonnet 4.6 for complex analysis, and fast, cost-effective models like GPT-5.4 for simple routing.

  • Native support for OpenAI, Anthropic, Google Gemini, DeepSeek, Grok (xAI), Perplexity, and HuggingFace.
  • Local deployment support via Ollama for sensitive or air-gapped environments.
  • Dynamic model switching based on task complexity.
  • Unified credential management across all providers.
agent_config.jsonv1.2.0

"name": "DataAnalyzer",

"provider": "anthropic",

"model": "claude-sonnet-4-6",

"temperature": 0.7,

"tools": [

"postgres_query",

"slack_notifier"

],

"guardrails": "strict_compliance"

Supported Providers

Click any provider to view setup instructions, available models, and configuration details.

Common Agent Roles

While agents are flexible, we recommend organizing them into specialized functional roles.

Router Agent

Acts as a dispatcher. It analyzes incoming requests and routes them to the most appropriate specialized agent or workflow branch.

Executor Agent

Designed to take action. This agent has write access to tools and is responsible for executing tasks like database updates or API calls.

Critic Agent

A specialized quality-control agent that reviews the output of other agents against a set of predefined standards before finalization.

RAG Agent

Optimized for Retrieval-Augmented Generation. It has access to your vector databases and document stores to answer questions with factual grounding.

Ready to build your first agent?

Jump into the Workflow Designer and start configuring your agentic swarm.