The octopus architecture for AI agents

TorkBot uses an "octopus" architecture with a centralized LLM conversation (foreground lane) coordinating multiple semi-autonomous appendages (lanes) that handle specialized tasks, allowing the system to maintain continuous personality and context while delegating complex work. The design balances three competing pressures: responsive interactions through bounded turns, capability through task delegation, and continuity through a single unified LLM conversation across all platforms and surfaces. This approach bets that future AI models will be capable enough to understand priority, recency, and interruption across collapsed platform boundaries, enabling seamless work continuation across different services.