Context Warp Drive: deterministic folding for long-running LLM agents

Short summary

Context Warp Drive is a TypeScript library that keeps long-running LLM agents under the context ceiling through deterministic folding—compacting old transcript regions into stable artifacts while preserving exact recall coordinates and enabling provider prompt caching. Instead of summarization calls, it maintains byte-stable prefixes and keeps recent messages append-only. Open-sourced on GitHub with 459 deterministic tests, the author is seeking feedback on whether this lower-level abstraction is the right primitive for agent memory or if it belongs higher in the application stack.

• •Deterministic folding compacts old transcripts into stable fold artifacts without summarization calls
• •Byte-stable prefixes enable provider prompt caching (Anthropic, OpenAI, Gemini)
• •459 tests published; npm release pending auth; author seeking design feedback on primitive shape

Generated with AI, which can make mistakes.