caura prism · design preview
Explore a proposed way to see a tenant’s memory as a field and connect stakeholder questions to actions: prune, explain, resolve, retune, re-scope, tune. The views below illustrate the design direction.
Design preview with synthetic data. Current Prism includes a memory/document browser; additional views depend on organization features. The eight-lens field design shown here is proposed.
Design preview with synthetic data. Current Prism includes a memory/document browser; additional views depend on organization features. The eight-lens field design shown here is proposed.
Proposed field: memories placed by meaning, with brightness representing recall.
Proposed risk halos: findings linked to a memory, its writer and its scope.
Illustrative design for visibility, cost, audit and a daily brief.
Design concepts for stakeholder questions and actions, illustrated on synthetic data. These cards describe the proposed field console; they are not a list of currently available product views.
proposed · field · platform lead
Every memory as a point, placed by meaning, bright where it is recalled and dim where it is not. Dense regions are what the fleet knows well; empty ones are gaps.
proposed actionPrune what nobody recalls; seed what is missing.
proposed · recall · auditor
Replay any recall: the question, the candidates, how each ranked and why, and which memories the answer was built from.
proposed actionExplain a decision to a reviewer, with the evidence attached.
proposed · agents · head of ai
Memory territories by writing agent, and which memories crossed from one agent to another this week. What never moved is visible too.
proposed actionRetune agents that hoard, or ones that never recall.
proposed · evolve · knowledge owner
Contradictions between memories, facts that were superseded, and what has gone stale, each with its provenance.
proposed actionResolve the conflict; retire the old fact; keep the history.
proposed · spend · finance
Lifetime tokens per memory: what it cost to write, what it earned in recalls, cold memory never recalled, and tokens avoided. The assumption behind every estimate stays on screen.
proposed actionTune retention and recall depth per agent; showback per team.
proposed · risk · ciso
Every credential, key or PII finding as a red halo on the memory that carries it, tied to the agent that wrote it and the scope it sits in.
proposed actionRe-scope, mask or drop it; rotate the secret.
proposed · audit · compliance
Every read, write and delete, attributed to an agent, including recalls that returned nothing.
proposed actionAnswer an auditor from the log, not from memory.
proposed · daily brief · everyone
A short brief written for people: what the fleet recalled and decided, what it delivered, and what needs a human's attention.
proposed actionRead it in two minutes; act on the third line.
Synthetic design sample: 2,500 memories, 5 agents.
43.78M
tokens avoided, assuming one in four recalls replaces a source re-read. The assumption stays on screen.
8%
cold memory: 206 memories stored and never recalled. Review candidates, not automatically waste.
37%
reuse rate: durable memories recalled by two or more agents.
117
memories reused across agents this week, each with who wrote it and who recalled it.
design sample · a proposed daily brief, written for people
1,214 recalls · 41 decisions recovered by the Interviewer · 27 outcomes reported back.
A duplicate refund traced to a billing-job retry and reversed the same day.
Rotate the API key pasted in #billing-ops. Remove the service-account JSON from a postmortem.
A fleet-memory pilot uses currently available product capabilities. Discuss the proposed Prism views and which measures matter for your workflow.
Plan a pilot