Packet28 – Context Layer for AI coding agents
Slash agent token costs 164x with hook reducers and persistent daemon state.

Persistent context layer beats Cursor's session amnesia on large codebases.
Engineering teams using AI coding assistants
Cursor · Sourcegraph Cody · Continue
Slash agent token costs 164x with hook reducers and persistent daemon state.
Cross-provider agent memory is clever, but LLM context windows keep growing and RAG is already standard.
The describe → plan → act split is an elegant, accessibility-inspired way to give LLM agents actionable UI context: annotate with data-ai-* attributes or use the Marker component, call describe(), send it to a planner, then client.act() executes DOM instructions. It's a clever middle layer that turns messy DOM state into structured inputs for server-side planning, though adoption will hinge on robust selector semantics and out-of-the-box integrations with popular LLMs and automation backends.
Three-method API for agent memory, but semantic memory systems aren't novel anymore.
Human-curated context beats auto-RAG, but folders-as-context is a solved workflow pattern.
Decision memory with enforceable context beats Cursor's built-in context features.