Build
Define agents and configure models, tools, context, and StreamNative Cloud integrations in Orca Agent Engine.
Building in Orca Agent Engine starts with configuration, not deployment. An Agent is a versioned registry resource that declares the model to run on, the system prompt that frames behavior, the tools the agent can call, and the context it carries: skills, memory, files, and repositories. On StreamNative Cloud, this tab also covers the Workspace integrations around agents - connections, packages, functions over Pulsar or Kafka, sources, sinks, and Kafka Connect.
Starting an agent, streaming its output, and supplying credentials at run time are on Run.
Agent
Define agents
Create, version, and manage the Agent resource that everything else hangs off.
Model access
Choose the model an agent runs on and bind the provider credentials behind it.
Model pricing
Inspect effective model prices used for cost accounting and budgets.
Harnesses
The loop that runs the agent. The harness decides where tool calls execute and which parts of the session you can drive, so it is chosen per agent rather than per session.
Claude Agent SDK harness
The default harness: full tool set, threads, and client-executed tools.
Claude Code harness
Runs the loop inside the sandbox, and what that costs you.
Codex SDK harness
Run OpenAI models with the managed Codex SDK in either execution mode.
Pi SDK harness
Run supported models from multiple providers with the managed Pi SDK.
Tools
What the agent can do beyond producing text. Tools come from three places: a built-in toolset that works on the session sandbox, MCP servers you declare, and custom tools your own application runs.
Tools
Declare the built-in sandbox toolset and custom tools your application executes.
Govern tool use
Decide which tool calls run, pause for approval, or are denied, with permission policies and guardrails.
MCP servers
Declare Model Context Protocol servers for the default harness, which routes their calls through the gateway.
Cloud extensions
StreamNative Cloud adds Workspace integrations under the cloud.sn.io/v1 extension group. The CLI
and TypeScript SDK discover the group before using these resources; a self-hosted deployment does
not serve them.
Connections
Configure reusable Pulsar, Kafka, and generic endpoints.
Packages
Store versioned function, source, and sink artifacts.
Functions
Run lightweight stream processing over Pulsar or Kafka alongside agents.
Sources
Ingest external data into Pulsar topics.
Sinks
Deliver Pulsar records to external systems.
Kafka Connect
Deploy and operate Kafka Connect plugins and connectors.
Context
What the agent knows, and what it can read or write while it works.
Skills
Package reusable, filesystem-shaped capabilities that agents load on demand.
Memory
Persist what an agent learns across sessions, with versioning and redaction.
Files
Upload assets and mount them as resources inside a session.
GitHub
Mount a repository into the sandbox and give the agent GitHub tools.