Metadata-Version: 2.4
Name: deepkeel
Version: 0.0.1
Summary: A context keel for AI agents — invisible, structural, stabilizing. (Name reservation; the real release follows.)
Author: Wukong
License: Apache-2.0
Project-URL: Homepage, https://github.com/cedricwyh/deepkeel
Keywords: llm,agents,context,compaction,kv-cache
Classifier: Programming Language :: Python :: 3
Classifier: License :: OSI Approved :: Apache Software License
Classifier: Operating System :: OS Independent
Requires-Python: >=3.10
Description-Content-Type: text/markdown

# DeepKeel

**A context keel for AI agents — invisible, structural, stabilizing.**

> This release is a **name reservation** (`0.0.1`). The real package is in
> development; the API published today is a placeholder and will be replaced.

DeepKeel treats the context window as an auditable ledger of structured
segments: it governs layout, budget allocation and tiered compaction *below the
waterline*, where the agent cannot see it — the way a keel holds a ship steady
without ever breaking the surface.

- Keeps prompt prefixes byte-stable so provider caches keep hitting
- Prunes oversized tool receipts without touching the user's instructions
- Compacts in tiers (prune → dedupe → summarize) and reports what it dropped

Repository: https://github.com/cedricwyh/deepkeel
