Core capabilities
These five capabilities form the product circuit every other Zephr capability consumes: memory keeps the belief base, retrieval reads it, trust decides what may be disclosed, code intelligence grounds beliefs in verifiable structure, and continuity moves working state between sessions, tools, and machines.
Storage topology, tenancy, and deployment boundaries are described once in Architecture and are not restated here.
Pages
| Page | Status | In one line |
|---|---|---|
| Memory | evolving | Write path and lifecycle of the belief base (L0–L2 shipped; L3–L5 planned) |
| Retrieval | evolving | The memory read path (local + hosted fuse) with fusion, calibration, abstention, and a CI-enforced parity gate |
| Trust | evolving | Review state machine, provenance, and the Trust Firewall |
| Code intelligence | evolving | Six retrieval tiers, query routing, and freshness |
| Continuity | evolving | Signed handoff packets that survive tool, machine, and vendor switches |
Status honesty
The baseline rows memory, retrieval, trust, code-intelligence, and
continuity are tracked in openspec/CAPABILITIES.md (Core /
Differentiator classes); each page above states what is shipped versus
specified. Tool names, tiers, and residency in these pages mirror the generated
MCP tool reference and
CLI reference, which are regenerated
from the shipped registry rather than hand-maintained.