One app, every AI provider
Claude · Codex · Copilot · Antigravity · OpenCode · Aider · Grok · Mistral Vibe · Kimi · Amp · oh-my-pi · Freebuff · Devin · Cursor

AgentsRoom supports 14 AI coding providers: Claude Code (Anthropic) with Opus, Sonnet and Haiku models; Codex CLI (OpenAI) with the GPT-6 / GPT-5.6 family (Astra, Sol, Sol Max, Terra, Luna); GitHub Copilot CLI (GitHub) with Claude Sonnet 4.5, GPT-5 and Claude Haiku 4.5 on your existing Copilot seat; OpenCode with Claude Sonnet, GPT-5 and Claude Opus; Antigravity CLI (Google) with Gemini 3.1 Pro and Gemini Flash; Aider with Claude Sonnet, Claude Opus, GPT-4o and DeepSeek; Grok Build (xAI) with Grok 4.6 and Grok 4.5; Mistral Vibe (Mistral AI) with Mistral Medium 3.5 and Devstral; Kimi Code (Moonshot AI) with Kimi K3 and Kimi K2.7 Code; Amp (Sourcegraph), which resolves its own model behind low/medium/high/ultra modes instead of a model picker; oh-my-pi, an open-source agent routing to 60+ model backends; Freebuff, a free agent built on the Codebuff platform with no model picker; Devin (Cognition), whose single plan reaches Claude, GPT, Gemini, Cognition's own SWE models and open-source ones like GLM, Kimi and DeepSeek, and can hand a session off to a cloud agent; and Cursor (Anysphere), the terminal agent of the Cursor editor, whose plan reaches Anthropic, OpenAI and Google models plus Cursor's own Composer, with the catalogue read from your own account at pick time. GitHub Copilot CLI here means the standalone `copilot` agent from npm (@github/copilot), not the gh copilot shell-command helper.

Yes. AgentsRoom builds a handoff context when you switch providers. It includes the list of modified files, session activity, and optionally a detailed summary written by the current agent. The new provider receives all of this as its first prompt and picks up exactly where the previous agent left off.

Absolutely. Each agent in AgentsRoom has its own provider and model setting. You can have one agent running Claude Opus for complex architecture, another running Codex in full-auto mode for repetitive tasks, and a third running Antigravity CLI on the free tier, all in the same project.

Yes. On the mobile companion app (iOS and Android), long-press any agent to open the provider switcher. Select a new provider and model, and the switch command is sent to your desktop via the end-to-end encrypted relay. The switch happens on your machine.

Yes. When both the old and new provider are Claude, AgentsRoom sends a live model change command directly to the running session. There is no restart, no kill, no handoff. The model switch is instant and the conversation continues uninterrupted.

Yes. Each agent has a per-agent CLI options field that passes flags straight to the underlying CLI at launch. Add a model flag such as --model opus-4.8 to pin a specific version, for example an Opus release your plan still has access to, without waiting for it to appear in the model picker. It overrides the provider-wide CLI options set in Settings, and applies to that agent only. When your flags already pin a model, AgentsRoom adds none of its own, so the CLI only ever sees one: that is how you reach a model alias served by an external OpenAI-compatible gateway. Picking Default in the model picker has the same effect with no flag at all, leaving the CLI on its own model.

Yes. OpenCode Go is a low-cost flat-rate plan that bundles a curated set of open models behind one monthly price instead of per-token billing. Connect it from inside OpenCode with the /connect command: its opencode-go models are then discovered from your own CLI and listed in the AgentsRoom model picker, so an agent can run on them without you holding any provider key. Its endpoint also speaks the OpenAI and Anthropic formats, so you can point a Claude Code or Codex agent at it with two environment variables in Settings, and keep your premium subscriptions for the agents that really need them.