Skip to main content
Runtime supports seven coding agents. Every session and background agent run is powered by one of these. Pick the agent that fits your workflow, or run different agents in parallel across sessions.

​
Prompt API agents

These agents are orchestrated by Runtime’s prompt API. You send prompts via REST or WebSocket and get back structured output. Runtime handles streaming, error recovery, and multi-step workflows.
Anthropic · claude-code · docsThe most full-featured agent on Runtime. Supports plan mode, real-time streaming, task tracking, auto-fix on errors, and session continuity across prompts. The default agent for new sessions.Models: Sonnet 4.6 (default), Opus 4.7, Haiku 4.5Key setup: Provide your Anthropic API key on the Integrations page, or use Runtime credits.
OpenAI · codex · docsOpenAI’s coding agent for autonomous development. Runs via the prompt API with auto-fix support.Models: GPT-5 Codex (default), GPT-5, o4-miniKey setup: Provide your OpenAI API key on the Integrations page.
Open source · opencode · docsOpen-source coding agent with broad model support spanning both Anthropic and OpenAI providers. Runs via the prompt API with auto-fix support.Models: Sonnet 4.6 (default), Opus 4.7, Haiku 4.5, GPT-5, o3, o4-miniKey setup: Provide your Anthropic or OpenAI API key on the Integrations page.

​
Interactive CLI agents

These agents run their own command-line interface inside the session terminal. You interact with them through the Terminal WebSocket or the dashboard’s embedded terminal. Runtime handles sandbox setup, API key injection, and lifecycle management.
GitHub · github-copilot · docsGitHub’s AI coding assistant with CLI integration. Authenticates automatically via your GitHub token.Models: GPT-5, Claude Sonnet 4.6Key setup: Provide a GitHub personal access token (GH_TOKEN) on the Integrations page. Get one at github.com/settings/tokens.
Cursor · cursor-cli · docsCursor’s terminal-native coding agent. Runs interactively inside the embedded terminal.Key setup: Provide your Cursor API key (CURSOR_API_KEY) on the Integrations page. Get one at cursor.com/dashboard.
Cognition · devin-cli · docsCognition’s autonomous software engineer. Runs interactively with its own CLI flow inside the terminal.Key setup: Provide your Devin API key (DEVIN_API_KEY) on the Integrations page. Get one at app.devin.ai/settings/api.
Google · gemini-cli · docsGoogle’s Gemini coding agent for the terminal. Runs interactively with its own CLI.Key setup: Provide your Gemini API key (GEMINI_API_KEY) on the Integrations page. Get one at aistudio.google.com.

​
Capabilities at a glance

​
Choosing an agent

Specify the agent when creating a session or launching a background agent:
Valid values: claude-code, codex, opencode, github-copilot, cursor-cli, devin-cli, gemini-cli. If omitted, defaults to claude-code.

​
Where agents run

Coding agents power two Runtime workflows:
  • Sessions - interactive environments where you work alongside the agent in real time, sending prompts and watching output
  • Background agents - autonomous runs triggered from Linear, Slack, GitHub, or the API that produce a pull request without human intervention
The same agent, same sandbox, same context and guardrails apply in both cases. The difference is whether you’re watching or not.