What is Octavus - Declarative AI Agent Orchestration Platform

Octavus is a declarative AI agent orchestration platform for developers. It provides a structured protocol for defining agent behavior, an execution platform that handles runtime concerns, and open-source TypeScript SDKs for server and client integration.

Overview

Octavus enables developers to build, deploy, and manage AI agents without writing orchestration infrastructure. Agents are defined declaratively — specifying their model, system prompt, available tools, triggers, and handler workflows. The Octavus platform then handles model routing, session management, SSE streaming, tool execution coordination, and execution tracing.

The platform is model-agnostic, supporting agents powered by models from Anthropic (Claude), OpenAI (GPT), Google (Gemini), and other providers. Developers can use different models for different steps within a single agent workflow, and switch models without modifying code.

Architecture

Octavus uses a three-layer architecture:

Agent Protocol — A declarative definition of agent behavior. The protocol specifies triggers (how the agent is invoked), handlers (what steps to execute), tools (capabilities available to the agent), and configuration (model, system prompt, parameters). Protocols are structured, readable, and version-controllable.
Orchestration Platform — A managed runtime that executes agent protocols. The platform handles model API calls, session state persistence, real-time SSE streaming, tool call coordination, and execution tracing. Developers interact with the platform through the Octavus web interface or the CLI.
Developer SDKs — Open-source TypeScript packages for integrating agents into applications. The server SDK handles tool execution on the developer's backend. The client SDK and React hooks handle frontend integration with real-time streaming.

Key Capabilities

Declarative Agent Definition

Agent behavior is declared in a protocol rather than coded imperatively. This separates what an agent does from how it's executed. The protocol is readable by non-engineers, version-controllable, and portable across environments.

Model-Agnostic Orchestration

Agents can use any supported model provider. Models are specified as provider/model strings (e.g., anthropic/claude-sonnet-4-5, openai/gpt-4o). Different models can be used for different steps within the same handler. Switching providers is a configuration change, not a code change.

Session Management

The platform manages conversation sessions automatically. Sessions persist across page refreshes and network interruptions, handle context window management, and support restoration for continuing conversations.

Real-Time Streaming

All agent interactions support SSE (Server-Sent Events) streaming. The streaming protocol is compatible with the Vercel AI SDK, enabling progressive rendering of responses, tool call progress, and structured events.

Tool Execution on Developer Infrastructure

Tools are defined as typed contracts in the protocol and executed on the developer's backend through the server SDK. The platform coordinates tool calls but does not execute them directly, keeping sensitive data under the developer's control.

Agent Observability

Every execution step is automatically traced: model calls, tool invocations, reasoning steps, timing data, and skill activations. Traces are available in real-time through the platform interface.

Agent Skills

Agents can be extended with knowledge packages following the Agent Skills open standard. Skills are provider-agnostic and work with any LLM, allowing developers to add specialized capabilities to agents without modifying the base protocol.

SDK Packages

The Octavus JavaScript SDK is open source on GitHub. Available packages:

Package	Description
`@octavus/core`	Shared types and utilities
`@octavus/server-sdk`	Server SDK for backend integration and tool execution
`@octavus/client-sdk`	Framework-agnostic client SDK for connecting to agents
`@octavus/react`	React hooks and bindings for real-time agent interaction
`@octavus/cli`	CLI for agent management and deployment

How It Compares

Octavus differs from other agent frameworks primarily in its declarative architecture. Unlike imperative frameworks (such as LangChain) where developers wire together chains, memory classes, and tool executors in code, Octavus developers declare agent behavior and the platform handles execution. Unlike role-based frameworks (such as CrewAI) that simulate multi-agent teams, Octavus focuses on single-agent workflows with powerful tool and handler capabilities. Unlike provider-specific SDKs (such as the OpenAI Agents SDK), Octavus is model-agnostic and works across providers.

For detailed comparisons, see Octavus vs LangChain and Octavus vs CrewAI.

Getting Started

The Octavus documentation covers installation, protocol definition, SDK integration, and deployment. The platform offers a free tier for development and testing.

What is Octavus?