Ollama just gave Claude Code a second brain for parallel tasks

Ollama shipped a significant update to its Claude Code integration, adding native subagent orchestration and built-in web search. Developers can now spin up parallel agents for tasks like code exploration, security audits, and real-time research. No MCP servers or external API keys needed. For a deeper look at how parallel agents change coordination, see the analysis of parallel agent coordination.

According to the company's announcement, the subagent system runs side tasks in their own context windows. This keeps the main conversation clean and lets long coding sessions stay productive, exploration and debugging work no longer fills the model's active memory. The approach echoes a pattern also seen in OpenManus's agent framework.

Some models can trigger subagents on their own when they spot a task that would benefit from parallel work. Those models are minimax-m2.5, glm-5, and kimi-k2.5. Developers can also force the behavior by explicitly telling the model to "use subagents" or "spawn subagents." Notably, MiniMax's M2.7 has shown strong coding benchmark results.

Example prompts include spawning agents to explore authentication flows, payment integrations, and notification systems all at once, or running audits for security issues, performance bottlenecks, and accessibility in parallel.

Web search built into the compatibility layer

Ollama's web search capability lives inside the Anthropic compatibility layer. When a model needs current information, Ollama handles the query inline and returns results without extra configuration. This simplicity stands in contrast to the deeper traps that can catch production agents, the subtle failures that emerge in production.

Subagents can use web search independently too. A single command like "research postgres 18 release notes, audit our queries for deprecated patterns, and create migration tasks" launches three research agents. Each fetches live data and returns actionable results.

Ollama recommends three cloud models for the new features: minimax-m2.5:cloud, glm-5:cloud, and kimi-k2.5:cloud. Start with ollama launch claude, model minimax-m2.5:cloud. This update positions Ollama as a lightweight alternative to heavier agent frameworks, a bet that aligns with the company's broader strategy, as its $88 million raise suggests.

The update positions Ollama as a lightweight alternative to heavier agent frameworks. Built-in parallelism and web access without the infrastructure overhead.