openai client libraries work with Orca. You find it under client.beta.agents.
Artifact: A copy of a file the agent wrote to its outputs folder, saved when a turn completes. Artifacts do not change after they are saved. See files and artifacts.
Base URL: https://api.orcapods.ai/v1, the address of the Orca API. Pass it as base_url (Python) or baseURL (TypeScript).
Cancel: Stopping a running turn by sending an agent.session.input.cancel event. Closing a stream does not cancel. See streaming and cancellation.
Credential: One secret stored in a vault, such as a bearer token for one MCP server URL. Orca never returns it after you store it. See credentials and vaults.
Environment: The machine where the agent runs commands and works with files. One of none, openai_hosted (a managed sandbox), or self_hosted (your own machine, coming soon on hosted Orca). See execution environments.
Environment template: A reusable recipe for a managed sandbox: files to copy in, skills to install, network policy. See skills and templates.
Event: A message on a session’s live stream (output) or a request you post to a session (input). Output events show progress; input events send messages, tool results, and cancellations. See events reference.
Executor: The program that connects a self-hosted machine to Orca. Orca uses the stock Codex exec-server. See self-hosted executor.
Function tool: A tool that your application implements. When the model calls it, Orca pauses the turn until your code sends back a result. See function tools.
Idempotency key: A unique value you send in the Idempotency-Key header so that retrying a request does not do the work twice. See reliable session workflows.
Item: One entry in a session’s saved history: a message, a tool call, a tool result, a command, and so on. See how Orca works.
Kit: A published bundle of agents, skills, and environment templates that anyone can read and copy. See kits.
Machine time: The time your sessions’ managed sandboxes run, counted in minutes. Each plan includes some. See plans.
Managed sandbox: An openai_hosted environment. Orca creates it for the session and deletes it with the session. Orca manages it; nothing runs at OpenAI.
MCP (Model Context Protocol): An open protocol for exposing tools over HTTP. An MCP server offers tools; Orca connects to it and lets the agent call them. See MCP tools.
Orca credit: The prepaid balance in your wallet. It pays for models named orca/openrouter/<id> and their web searches, and extra machine time. See wallet and credit.
Organization: The account everything in Orca belongs to: agents, sessions, keys, the wallet, and the plan. Your API key acts in one organization, and you only see its resources. See organizations and roles.
Plan: Free, Pro, or Max. It sets the included machine time and how many sessions run at once. See plans.
Provider key: Your own API key for a model provider such as Anthropic or OpenRouter, stored in Orca so your agents can use that provider. See models and provider keys.
required_actions: The list on a session that says what the agent is waiting for: a function result, or a self-hosted executor connection. See turns and statuses.
Role: Admin or member. Admins also manage provider keys, vaults, billing, and every API key. See roles.
Sandbox: Informal name for an environment that can run code, either managed or self-hosted.
Session: One conversation with an agent, plus its environment and attached vaults. Keep its ID to continue the conversation. See sessions and turns.
Skill: A versioned folder containing a SKILL.md file of instructions (and optionally scripts or data) that teaches the agent a specific procedure. See skills and templates.
SSE (server-sent events): The HTTP streaming format Orca uses for events. The openai clients handle it for you.
Steering: Sending a message while a turn is running. The message joins the running turn instead of starting a new one. See turns and statuses.
Subagent: A helper conversation that an agent starts to handle part of a task. Its result goes back to the parent agent. See subagents.
Tenant: The API’s word for your organization.
Terminal status: A turn status that means the turn is over: completed, failed, or cancelled.
Turn: All the work the agent does in response to one message, from the first model call to the final answer. See turns and statuses.
Vault: A container for credentials that a session’s MCP tools can use. The agent uses them without ever seeing them. See credentials and vaults.
Wallet: Your organization’s Orca credit balance. See wallet and credit.
Workspace: The working folder inside an environment. In a managed sandbox it is /workspace; in a self-hosted environment it is the workspace_directory you choose.