Gemini CLI: Remote Agents & A2A Multi-Agent Support
Gemini CLIGemini CLI v0.24.0 introduced support for remote agents and multi-agent TOML configuration files, enabling the CLI to delegate tasks to external specialized agents using the Agent-to-Agent (A2A) protocol. Developers can declare a network of remote agents in TOML files under .gemini/agents/, and the orchestrator routes subtasks to the appropriate remote agent based on capability matching. This positions Gemini CLI as an orchestration layer for multi-agent systems rather than a single-model tool.
Sources & Mentions
4 external resources covering this update
Remote Agents and Multi-Agent TOML Support Arrive in Gemini CLI v0.24.0
The v0.24.0 release of Gemini CLI, published January 14, 2026, marks the beginning of a significant architectural expansion: the CLI can now coordinate with remote agents using the Agent-to-Agent (A2A) protocol, transforming it from a standalone terminal assistant into an orchestration hub for distributed multi-agent workflows.
The Agent-to-Agent (A2A) Protocol
A2A is an open interoperability protocol for agent communication that Google co-developed and subsequently transferred to the Linux Foundation, where it is rapidly becoming an industry standard for agent-to-agent coordination. By building A2A support directly into Gemini CLI, Google positions the tool to interoperate with any A2A-compliant agent — whether a specialized coding agent, a database query agent, a testing agent, or a custom enterprise service.
When Gemini CLI receives a task that matches a remote agent's declared capabilities, the orchestrator delegates the subtask to that agent, receives the result, and incorporates it into the main response or plan. From the user's perspective, the interaction remains a single unified session; the multi-agent coordination happens transparently under the hood.
TOML-Based Agent Configuration
The v0.24.0 release introduces a declarative configuration model for defining agent networks. Developers place TOML files inside .gemini/agents/, each describing a remote agent's endpoint, capabilities, and trust level. The CLI scans these files at startup to build an agent registry — a fixed graph of available agents for the session.
This approach keeps agent configuration auditable and version-controllable: the agent topology lives in the repository alongside the code, making it straightforward for teams to review and reproduce the agent environment in CI/CD or collaborative settings. It also means that peer review processes can catch agent configuration changes before they reach production.
Model Routing and Silent Delegation
The v0.25.0 release built directly on this foundation by adding model routing to subagents — the ability to specify which underlying model powers each remote agent. A team might configure one agent to use a lightweight, fast model for simple retrieval tasks and another to use a more powerful model for complex reasoning. The orchestrator handles routing transparently based on task type.
Silent local delegation was also introduced, allowing lower-stakes subtasks to route to local agents without surfacing a confirmation dialog, while remote or higher-risk delegations continue to require explicit user confirmation. This balance between automation and oversight reflects the growing maturity of Gemini CLI's agentic model.
Enterprise and Registry Integration
The v0.24.0 release also added remote agents to the agent registry and introduced remote admin settings that let enterprise administrators control which remote agents are available to users, define trust levels per agent, and restrict delegation to an approved allowlist. Combined with the generalist agent added in v0.26.0 — which handles broader task orchestration when no specialist agent is defined — the multi-agent system now scales from simple two-agent setups to complex enterprise agent graphs.