What CacheTank does
CacheTank stores your context as structured text in Firestore. When any AI reads your context link, it gets exactly what you saved — assembled into a clean briefing. No embedding model decides what's "relevant." No vector similarity search guesses what you meant. No LLM rewrites your context before the AI sees it.
What competitors do
Every competitor processes your context through AI before serving it. ctxvault uses ChromaDB vector store + sentence-transformer embeddings. XTrace uses encrypted vectors + semantic retrieval. Mem0 uses vector embeddings + LLM classification. Qorbit uses knowledge graph traversal. AI Context Flow uses AI to optimize your prompts.
Why this matters
When AI processes your context, it can lose nuance, hallucinate memories, or silently drop information it deems irrelevant. With CacheTank, what you saved is what gets served. Every time. You can read the output yourself and verify it — try doing that with a vector embedding.
How we verified this claim
Our backend source code (main.py, 1,787 lines) contains zero imports for embedding libraries, vector databases, or LLM APIs. The _read_context() function performs Firestore queries and assembles markdown. The extract_keywords() function uses basic string splitting. The wisdom promotion system uses set intersection — not cosine similarity.