Tools for coherent LLM coding agents.
We build infrastructure that gives AI coding agents persistent memory and orchestrated multi-agent coordination — modeled on how biological minds actually work.
reverie
daemonA memory consolidation daemon for LLM coding harnesses. Five-layer hierarchy, offline dream cycles, hybrid search — modeled on biological sleep consolidation.
meshctl
cli + tuiA seven-role mesh for orchestrating parallel Claude agents. One anchor dispatches, workers build in isolated worktrees, and meshctl holds it together with live telemetry.
Why this exists
LLM coding harnesses — Claude Code, Cursor, Windsurf — accumulate knowledge across sessions but lack a theory of where it should live. The default "dump everything into one store" degrades over time through duplication, misplacement, noise, and staleness. Meanwhile, running more than one agent at once requires ad-hoc tmux hacks that break the moment two sessions race on the same file.
Both problems have biological precedents. Memory consolidation is a 70-year-old area of neuroscience; multi-agent coordination is a textbook chapter in distributed systems. cerebral.work ports what's been learned in each into the daily workflow of building software with AI.