StreamNative Orca Agent Platform (Private Preview)
Deploy and manage AI agents that process real-time data at scale using industry-standard frameworks and models.Prerequisites
- StreamNative Cloud account with Orca Private Preview access enabled
- Basic understanding of event streaming concepts
- Familiarity with Python and your chosen agent framework (Google ADK, OpenAI Agents SDK)
What is StreamNative Orca Agent Platform
StreamNative Orca Agent enables developers to deploy and operate advanced AI agents handling realtime events securely and at scale. It provides streaming infrastructure purpose-built for event-driven agent workloads, robust tools to enhance agent capabilities, and enterprise-grade controls for real-world production deployments. Orca Agent services can be used individually or in combination and integrate seamlessly with popular agent frameworks, including Google Agent Development Kit, OpenAI Agents SDK, and Langchain (coming soon), as well as any foundation model, giving you maximum flexibility. By removing the undifferentiated heavy lifting of building and managing specialized streaming infrastructure, Orca Agent accelerates your path from development to production for tackling realtime events. Orca Agent Engine brings autonomous agent workloads to StreamNative Cloud by combining AI-first capabilities with the platform’s event-driven infrastructure. It lets you deploy reasoning agents without abandoning topic-centric pipelines, so stream processing, tool calls, and LLM orchestration share the same operational surface.Orca Agent Platform Capabilities
The Orca Agent Platform delivers a comprehensive set of capabilities designed to simplify the deployment, operation, and scaling of real-time AI agents. From seamless integration with streaming data systems to dynamic workflow orchestration, advanced context management, and enterprise-grade monitoring, Orca Agent equips teams with the tools needed to build, run, and govern production-ready agents with confidence. These capabilities work together to reduce complexity, accelerate time-to-market, and ensure reliability across diverse environments. Platform highlights- Runs inside the managed Orca agent runtime, inheriting StreamNative Cloud scaling, tenancy, and topic-based connectivity.
- Supports both Google Agent Development Kit (ADK) agents and OpenAI Agents runners today, with a shared agent runtime interface for future frameworks.
- Normalizes tool access through a managed context layer, allowing Model Context Protocol (MCP) toolchains to be injected at runtime alongside user-defined tools.
Real-Time Data Integration for Agents
Orca Agent Platform natively integrates with high-throughput messaging systems such as Apache Pulsar and Kafka, enabling agents to consume, process, and act on live event streams with ultra-low latency. This seamless connection allows agents to respond to changing conditions instantly, enrich decision-making with the latest information, and power real-time applications at any scale. Because data ingestion, routing, and back-pressure management are built into the platform, teams avoid the complexity of building custom pipelines or maintaining additional middleware. Native Pulsar and Kafka adapters integrate with Schema Registry across StreamNative Cloud clusters. Schema handling is unified so agent functions can ingest or emit different schema payloads without custom serialization code.Workflow Orchestration and Agent Coordination
The platform provides flexible orchestration support that allows you to define, schedule, and coordinate agents and workflows around your business logic. You can chain agents together, run them in parallel, or dynamically adjust execution paths based on rules or events, enabling complex, event-driven architectures. This orchestration capability reduces operational overhead and allows organizations to adapt quickly to new requirements or scale up specific functions on demand.Runtime Architecture
- Runtime interface provides async-friendly initialization and processing hooks, so agent loops can await external LLM calls without blocking the platform runtime.
- Context layer wraps the execution environment, exposing messaging clients, tool registries, and session mode so agent code can remain framework-agnostic.
- Base runtime loads agent packages, wires Google ADK or OpenAI runners, and reuses the built-in artifact service for intermediate outputs.
Session and State Management
- The runtime session service persists agent state in the managed store and streams events to per-session topics.
- Multiple session modes (
shared
,session_per_message
,session_per_user
) let you decide whether conversations share context or stay isolated; select the mode that fits your workload during agent configuration. - Producer instances are cached so the platform avoids recreating Pulsar or Kafka producers on every turn, keeping latency low for conversational workloads.
Enhanced Agent Functionality with MCP and Context Management
With built-in support for MCP tool calling, agents can securely invoke external services, APIs, or data sources to extend their functionality far beyond the core model. Coupled with first-class memory and context management, agents gain the ability to maintain state, retain knowledge across sessions, and personalize responses or actions based on historical interactions. This combination transforms agents from stateless responders into context-aware, adaptive actors capable of executing sophisticated workflows and reasoning over time.- Google ADK agents can load Python modules, merge MCP-provided tools with user tools, and run through the ADK runner with optional persistent sessions.
- OpenAI agents execute configurations built with the
openai-agents
package, maintain topic-based session caches, and reload MCP tools each turn. - MCP tool merging prevents duplicate registrations: both adapters filter out tools that already come from the managed tool registry before assembling the final tool list for the framework runtime.
- Support for MCP server-sent events (SSE) and streaming responses keeps agents in sync with external tools through real-time updates.
Enterprise-Grade Monitoring and Administration
Orca Agent Platform offers a unified observability and management layer, providing enterprise-grade monitoring, logging, and auditing across all deployed agents. Administrators can track performance metrics, resource utilization, and agent activities in real time, ensuring compliance and operational transparency. Fine-grained access control, secure isolation, and automated policy enforcement protect sensitive data and maintain consistent standards across multi-cloud or hybrid environments.- Agent admin API exposes REST endpoints aligned with agent management workflows, including listing, describing, creating, and deleting agents across namespaces and tenants.
- CLI tooling:
snctl
ships agent-focused commands that enable day-to-day lifecycle management (deploy, update, status, logs) from the terminal. - Cloud Console (UI) surfaces agent details in the Agents panel so teams can monitor deployments without leaving the browser.
Orca Agent Platform Benefits
Simplicity - Faster Development
- Python support as a first-class citizen: Build, deploy, and extend agents using Python with full native support, enabling developers to work with familiar tooling, libraries, and workflows.
- No extra infrastructure to spin up or manage: Orca Agent runs on a fully managed platform, eliminating the need to provision or maintain separate servers, containers, or clusters.
- No additional framework or API to introduce: Seamlessly integrate with your existing stack without learning new APIs or frameworks, reducing development overhead and accelerating time-to-market.
- Minimal code changes to existing agents: Existing agents can be onboarded with little to no refactoring, allowing teams to migrate quickly without disrupting production workloads.
Flexibility - Build Your Way
- Agent framework at users’ choice: Support for Google ADK, OpenAI Agents SDK, and other popular agent frameworks gives teams freedom to build with the tools they know best.
- Messaging protocol at users’ choice: Native integration with Pulsar or Kafka lets you connect to your preferred streaming backbone without additional middleware.
- Foundation model at users’ choice: Work with Anthropic Claude, Google Gemini, OpenAI GPT, and other foundation models to adapt quickly to new capabilities and use cases.
- Hyperscaler at users’ choice: Deploy across AWS, GCP, or Azure to take advantage of your existing cloud footprint and optimize for cost, performance, or compliance.
Security - Enterprise-Grade Protection
- Fine-grained data access control inherited from StreamNative Enterprise: Enforce enterprise-grade identity, authorization, and data access policies without additional configuration.
- Complete isolation between agents to minimize interference risk: Each agent runs in an isolated environment, preventing data leakage, cross-contamination, and performance impact from other agents.
- Ability to audit and trace agent activities whenever needed: Built-in auditing and observability allow you to track agent behaviors, ensure compliance, and quickly troubleshoot issues.
Cloud-Native - Operate at Scale
- Native auto-scaling capabilities: Automatically scale agents up or down in response to demand, ensuring high performance and cost efficiency without manual intervention.
- Strong fault tolerance with minimal maintenance overhead: Built-in redundancy and recovery mechanisms ensure continuous availability and reduced operational burden.
- Collaborate with other workflows/agents within seconds: Natively orchestrate or chain multiple agents and workflows together in real time, enabling complex, event-driven applications.
What’s next?
- Review stream processing fundamentals in Stream processing overview before deploying agents.
- Learn how to set up your environment.