Real Claude context-management concepts, in one durable catalog.
Every request to Claude shares one finite token budget across the system prompt, conversation history, tool definitions/results, and any attached documents -- the context window. This site catalogs the real mechanisms for working within that budget: prompt caching (reusing a cached prefix across requests for lower cost/latency) and context editing / compaction (trimming or summarizing older context to stay within budget).
Exact numeric token-limit figures are deliberately not stated as precise current facts unless independently verified -- this catalog phrases them generally rather than risk a stale or wrong number.
Prompt-caching rows cite this repo's own mirrored doc; other rows are phrased generally where an exact figure wasn't independently re-verified.
Token-limit specifics are described qualitatively rather than pinned to a number that could be wrong or drift.
Context windows, prompt caching, and context editing/compaction -- three real, distinct mechanisms.
Live catalog
All rows are seeded into this Worker's own D1 database (context_concepts). Browse by category → or hit the JSON API.