01
Evidence coverage
Claims, controls, obligations, and case decisions without support are visible as gaps.
Agent memory
Agent memory turns source material into context an agent can act on.
It is not a folder of documents or a search index. It is a governed layer of records, rules, evidence, state, permissions, and history. The agent gets the context it needs, along with the procedure for applying it and the proof for trusting it.
What it is
Operational memory for a non-deterministic actor.
A human can read a policy and infer the surrounding context. An agent needs that context externalized. Agent memory gives the model a governed structure around the work: what is known, which procedure applies, which source proves it, and whether the result is safe to use.
The model still reasons. The memory holds the load-bearing facts and procedures outside the model, where they can be reviewed, checked, cited, and updated without rewriting a prompt.
- Records The canonical facts, obligations, decisions, controls, commitments, and exceptions an agent can inspect.
- Rules Executable procedures that decide what applies, what passes, and what needs review.
- Evidence The source artifacts that justify the record: tickets, policies, contracts, files, transcripts, and approvals.
- State Whether a record is current, stale, disputed, superseded, reaffirmed, or waiting for review.
- Permissions Who may see, use, approve, or change the memory, carried with the record itself.
- History Every approved change, who made it, what source caused it, and what it replaced.
How it works
Memory is built from reviewed changes, not scraped documents.
The unit of change is a proposal against memory. Sources create proposals, people approve them, rules check them, and agents retrieve the approved context when they need to act.
- 01 Source
Work produces evidence.
A ticket closes, a policy changes, a call creates a commitment, or a control produces proof.
- 02 Proposal
WiseWare drafts a memory change.
The durable part is extracted as a record with sources, owners, dates, and rule context attached.
- 03 Review
A person approves the record.
AI can propose, but canonical memory changes only after review.
- 04 Check
Rules evaluate the result.
The approved record is checked for gaps, expiry, conflicts, permissions, and required evidence.
- 05 Return
The agent receives context.
Instead of a loose document match, the agent gets the relevant record, procedure, evidence, state, and caveats.
Memory health
The memory has to keep explaining why it should be trusted.
Agent memory is only useful while it remains inspectable, current, and tied to the rules that govern it. These are the signals WiseWare tracks so old context does not quietly become new instruction.
02
Rule status
Executable checks show what passes, fails, or needs human review.
03
Policy version
Records carry which policy, law, method, or obligation version applied.
04
Conflicts
Contradictions between records surface as disputes, not silent drift.
05
Review deadlines
Expired exceptions, stale evidence, and overdue approvals are marked before answers overclaim.
06
Ownership
Every operational record has an owner responsible for keeping it honest.
07
Permissions
Access rules travel with the memory, not just the source document or UI.
08
Supersession
When one decision, method, or policy replaces another, the old one points forward.
09
Version history
Every approved change is signed, readable, and reversible — the spine of trustworthy memory.
Agent memory is useful when it stays inspectable, current, and safe to execute.