Skip to content
GitHub

Runtimes

Compare the Claude Agent SDK, Classic, Anthropic and OpenAI Responses runtimes, and see which settings, channels and tabs each one offers.


On this page

Every agent runs on a runtime: the engine that executes its instructions, knowledge and tools. The runtime is chosen when you create the agent, and it decides which channels the agent can deploy to and which sections its tabs show. Use this page to pick a runtime, or to work out why a setting you expected is missing.

Choosing a runtime when you create an agent#

Open Agents and select Create Agent. The Create a new agent dialog asks for a name and then How should it work?:

  • Standard is the recommended slot. Runbear decides which runtime it creates for your organization: usually Claude Agent SDK, or Anthropic for organizations that Runbear has set up that way, including organizations with HIPAA mode. The card's description tells you which one you are getting.
  • Other runtimes is a disclosure under it. Expand it to choose Classic, the previous engine.

Press Command+Enter on macOS or Ctrl+Enter elsewhere to create the agent without reaching for the button.

Create manually, at the bottom of the dialog, opens the full creation form instead. Pick Anthropic Claude there for an Anthropic agent, or OpenAI for an OpenAI Responses agent. That form also lists older model providers, which this page doesn't cover.

"Standard" names only this slot in the create dialog. It is unrelated to standard data handling, the non-HIPAA setting in an agent's settings.

Capability matrix#

Claude Agent SDKClassicAnthropicOpenAI Responses
How to create itStandard slot (default)Other runtimes → ClassicStandard slot where Runbear has set it up, or Create manually → Anthropic ClaudeCreate manually → OpenAI
ChannelsSlack and Microsoft TeamsAll channelsAll channelsAll channels
Model sectionClaude, GPT and Gemini modelsNoneClaude modelsGPT models
Reasoning sectionNoNoNoYes
API Key section (your own key)Anthropic, OpenAI or Google AI Studio key, matching the selected modelNoAnthropic keyOpenAI key
MemoryFiles in the agent's workspaceRecall of related earlier messagesShared and personal factsShared and personal facts
Preferences tab on a chatbotHiddenShownShownShown
Activities tab on a chatbotShownShownShownHidden
Analytics tabHiddenShownShownShown
Slack triggers@mentions and DMs, AI-powered thread replies, Reply without @mention and Trigger by keywordAllAllAll
HIPAA data handlingNoNoYesNo
Classic Editor linkNoNoYesYes

The rows in detail:

  • Channels. An agent on the Claude Agent SDK runtime deploys to Slack and Microsoft Teams only; the Chatbots tab doesn't offer the other tiles. Your plan also limits which channels you can use; see channels by plan.
  • Model and Reasoning. See Model and Reasoning. Classic agents have no model setting; Runbear picks the model.
  • API Key. Lets the agent run on your own provider key instead of Runbear credits. See API key.
  • Memory. Each runtime remembers differently, and the Enable long-term memory toggle means something different on each. See Memory.
  • Preferences and Activities tabs. These are the per-connection tabs on the Chatbots tab. See Chatbot settings.
  • Analytics tab. The agent's own Analytics tab. Organization-wide analytics still include every agent.
  • Slack triggers. A Claude Agent SDK agent answers @mentions and DMs, can keep replying in a thread using AI-powered smart replies, can reply to every new channel message with Reply without @mention, and can reply to messages that contain a keyword. The other Slack triggers (schedules, emoji reactions and specific senders) need one of the other runtimes. See Slack triggers.
  • HIPAA. Only Anthropic agents can use HIPAA data handling. See the HIPAA setup guide.
  • Classic Editor. Anthropic and OpenAI Responses agents keep a link to the older editor at the bottom of the agent's sidebar. See the Classic Editor.