Slash Commands Simplify Provider Integration
Connect NVIDIA's API Catalog to Kilo CLI (or OpenCode fork) without editing configs, JSON providers, base URLs, or env vars. Get a free API key from build.nvidia.com by joining the developer program. In Kilo CLI, run /connect, select NVIDIA, paste the key—setup completes automatically. Then /models lists available options like Kimi K2.5, MiniMax M2.5, GLM-5. This one-time connection exposes multiple labs' models through NVIDIA, avoiding separate dashboards and billing. Free serverless access suits dev/testing but follows trial terms—not infinite production use.
Leverage Long-Context Models for Complex Tasks
Kimi K2.5 offers 256K token context as an open-source multimodal agentic model, ideal for retaining project state in multi-step coding. MiniMax M2.5 (204K context) excels at action-oriented tasks. GLM-5 (205K context) targets complex systems engineering and long-horizon agentic workflows with strong reasoning over large context. Access all via one provider, testing without per-token costs during dev.
Switch Models Mid-Workflow for Optimal Results
Post-setup, use Kilo CLI's agentic flow unchanged: inspect repos, analyze architecture, fix debt, build apps (e.g., Atari cropper, Next.js dashboard). Run /models to swap instantly—compare Kimi on one task, GLM-5 on reasoning-heavy refactors, MiniMax on long edits—without reconnecting. Test multiple prompts per model to match task styles. Caveats: Availability/limits may shift; verify /models list matches your NVIDIA catalog; free tier for testing, not heavy production.