Claude Code Ultraplan: 4x Faster Plans via Cloud Multi-Agents
Trigger Ultraplan in Claude Code CLI to offload planning to cloud agents on Opus 4.6, generating structured plans with diagrams in 1 minute vs 4+ minutes locally, leading to 3x faster execution and 38% fewer local tokens.
Trigger Ultraplan for Cloud-Powered Planning
Use /ultrplan or include "ultra plan" in your CLI prompt within Claude Code's terminal (not desktop app or VS Code extension) to send planning to Anthropic's cloud. It syncs your Git repo, scans the directory for existing files/skills, and builds a structured plan with tabs for context, new files, modifications, verification, and sometimes diagrams. Review on web with emoji reactions or section comments for iterative refinement—Claude re-plans based on feedback—then approve to teleport back to your local terminal for execution. This creates a richer review surface than terminal text, enabling targeted fixes like directing it to use specific skills (e.g., "visualizations" for Excalidraw diagrams instead of Mermaid).
Local planning blocks your terminal session; Ultraplan frees it, letting you multitask or start a new session while waiting (cloud cap: 30 minutes compute). Always git-init and push first—local-only projects fail with "can't do this on the web yet."
Superior Speed and Quality from Multi-Agent Design
Ultraplan uses Opus 4.6 (fixed model) with three parallel exploration agents plus one critique agent in Anthropic's cloud container runtime, versus local single-agent linear thinking on your session's model. In a dashboard build test (pulse board with MRR/ARR stats, revenue/cohort charts, customer tiers, support tickets, light/dark mode, time filters), local plan took 45 minutes total (12+ min planning, 30+ min execution, 131k tokens); Ultraplan planned in 5 minutes, executed in 5-10 minutes total (82k local tokens, ~50k extra cloud tokens estimated). Plans were comparably functional/aesthetic, but clearer structure sped execution—local struggled despite same prompt/model.
Other tests confirm Ultraplan plans faster and executes quicker locally due to precise structure. Abraham Lincoln quote applies: invest tokens upfront (e.g., API billing local: 1.5M input/23k output tokens; Ultraplan ~1% of 20x Max plan) for project success. Cloud compute outpaces local, yielding deeper plans without device limits.
Setup Requirements and Key Limitations
Requires Pro/Max subscription for cloud.web access; API billing sessions can't Ultraplan. Git sync essential for codebase awareness—without it, plans stay generic. Skills may need explicit prompting post-scan (e.g., comment "use visualizations skill" if overlooked). Random auth errors occur but retry fixes them. No token visibility for cloud phase yet; terminal unblocks but context may muddy if not using fresh session. Research preview status means evolving features like agent prompting details or token breakdowns incoming. Trade-off: higher upfront tokens/compute for 4x planning speed, better execution, structured web review—ideal for complex projects where planning dictates outcomes.