angelkurten/keel is a Python context layer that helps AI agents in vertical SaaS share customer context. The repository README reports the latest release as v0.26.0, and the available information does not list the specific changes in that version. The project takes domain events as input, builds a per-customer bitemporal context graph, and gives agents five MCP tools to read and write that context. The repository is licensed under Apache-2.0.
The core idea is that context is built from structured domain events. Each fact carries a source, a validity interval, a recorded time, and the subject that corrected it, so its history can be traced. It distinguishes not only the end customer but also that customer's own customers, tenants, with permissions and expiry periods set separately for each. Because several agents work with the same customer context, this differs from keeping separate information per agent. Search handles identifiers first and combines semantic search with lexical search, optionally reranking the results. Embedding models and stores can be swapped through adapters.
The README describes the project as being at alpha stage and states that it has fifteen Python test files. The available information includes no figures on performance or concurrency. What this repository offers is therefore an auditable context structure and a way to connect agents, not evidence of performance or scalability in production.