Kilo VS Code: Free Parallel AI Agents & Worktrees
Kilo's rebuilt VS Code extension shares CLI core for faster features, adds parallel tool calls/subagents, Git worktrees for isolation, and free access via Kilo/OpenRouter/NVIDIA models—turning it into a GA AI coding tool.
Shared Core Architecture Delivers Consistent, Fast Updates
Kilo rebuilt its VS Code extension on the same portable core as Kilo CLI, eliminating prior inconsistencies across CLI, JetBrains, and VS Code. This unification speeds feature delivery, improves performance, and simplifies maintenance—VS Code now acts as another frontend, ensuring parity. Result: New capabilities like parallel processing roll out everywhere simultaneously, avoiding the 'special snowflake' delays of the old VS Code-tied version.
Parallelism and Isolation Supercharge Agent Workflows
Parallel tool calls enable simultaneous file reads, searches, and commands, while parallel subagents delegate tasks—one implements features, another tests, a third reviews—boosting throughput over serial chats. Define custom subagents for team workflows. Agent Manager organizes multiple sessions for switching/comparing without tab chaos. Git worktrees isolate attempts in separate branches/workspaces, preventing conflicts during experimentation. Side-by-side comparisons test models/strategies directly; inline code review adds diff comments plus chat summaries, mimicking real dev reviews. Unified agents interface integrates sessions/reviews cohesively, with provider settings, MCP marketplace, session imports, dedicated terminals, and CLI/cloud sync for seamless cross-platform continuity.
Free Model Setup Unlocks Production-Ready Testing
Skip subscriptions: Use Kilo's built-in free-tier models (labeled 'free'). Connect OpenRouter API key for Qwen 3 Coder Free, GLM 4.5 Air Free, DeepSeek-R-10528 Free, Kimmy K2 Free—enable prompt training in OpenRouter settings if needed. For NVIDIA NIMs, use OpenAI-compatible provider: paste NVIDIA API key/base URL/model ID (e.g., Kimmy, GLM, MiniMax) for free developer access (testing terms apply, not infinite production). Pair with free Codestral (Mistral) autocomplete. All integrate into one workflow: select provider/model, leverage agents/worktrees/reviews consistently. April 2, 2026 update fills beta gaps, making it feel GA.