Every shipped feature, in depth: what it does, what it deliberately does not do, and where the
boundaries are.
This is the complete manual. The main README is the short version — enough to
install Knowl, understand why it retires stale knowledge, and get an agent talking to it. This
document is where every claim is spelled out, including the limits: which operations stay local,
which paths are lexical rather than semantic, what an import will and will not overwrite, and
which parts of a restore are deliberately not restored.
| Section |
Covers |
| Overview · Quick start |
What Knowl is and how to install it |
| Core knowledge model |
Atom categories · Metadata, history, ownership · Governed writes |
| Retrieval and context |
Current retrieval · What a result carries · Embedding models · Historical queries · Context packs |
| Tasks, sessions, lifecycle |
Work loops · Retention and promotion · Handoffs and resume keys · Transcript search · Host behavior |
| Evidence, code, drift |
Evidence and symbols · PR drift and feedback |
| Workspaces |
Federation and ownership · Reading a peer's atom by id · Doing a peer's work · Ownership stamping |
| Knowl Cloud |
Identity and connection · Publishing and drift · Staying current |
| Learned skills and synthesis |
File-backed skills · Deterministic synthesis |
| Portability and maintenance |
Export and import · Garbage collection · Snapshots, audit, doctor |
| Local viewer · Architecture |
Inspector, component diagram, security boundaries, write durability |
| Agent setup · Benchmarks |
Host integration, and every evaluation suite with reproduction commands |
| CLI reference · MCP tools |
Every shipped command, tool, and resource |
| Optional AI · Local data |
Provider configuration and on-disk layout |