Your council remembers everything across sessions — decisions, patterns, projects, open items. Built on a 4-tier system: segment summaries (mid-session), session summaries (daily), weekly digests, and a permanent milestone ledger. It runs on Sparks's Cloudflare infrastructure, so it persists even when your laptop is off. When you're ready, you can mirror it locally — but you don't need to.

90% of your council's work runs on Groq's Llama 3.3 70B — fast, free, and powerful. You start with Sparks's Groq allowance (~50K tokens/day). When you hit the limit you'll get a 5-minute prompt to set up your own free account. For private or offline work, your council can route to Sparks's Ollama instance (a local AI running 24/7 on Sparks's machine, tunnelled securely). Claude Sonnet/Opus is available when you need deep reasoning — you can add your own Anthropic key anytime, it stays entirely in your environment.

Every agent in your council can draw on a library of 90+ expert playbooks — agent orchestration, content strategy, filmmaking, marketing psychology, sales, code architecture, and more. These are SKILL.md files your agents read before producing output, making them domain experts on demand. You get the full library in your kit. When Sparks adds new skills, one git pull syncs them to your machine.

Type /kickoff and your Chief of Staff (Kairos) assembles the full council. They read your memory stack, check web context if needed, hold a structured boardroom discussion, and Kairos synthesizes. Everything is logged automatically — session summary, decisions locked, action register updated. You also get SOPs: how to run meetings, how to delegate, how to debrief, and the working patterns that make this system actually work at full power.