Research Concept

DomeAI

DomeAI explores persistent, searchable memory for AI agents, with explicit source context, access boundaries, and a reviewable record of what informed each action.

The current work focuses on the hard parts: retrieval quality, permission boundaries, stale context, and reliable handoffs between people and agents.

Shared Memory ConceptIllustrative
Scoped
Context
Linked
Sources
Reviewable
Actions
Proposed Flow

Record. Retrieve. Review.

01

Record

A proposed workflow records approved context with its source, owner, scope, timestamp, and retention rule.

02

Retrieve

Retrieval candidates would be ranked against a declared task while stale, conflicting, or unauthorized context remains visible.

03

Review

Retrieved context would support a prepared action only after source checks, permission checks, and the required human review.

Evaluation Scope

Questions a shared-memory system must answer

Persistence Boundaries

Evaluate which context should survive a session, how long it should remain, and when it must be removed or refreshed.

Scoped Context

Study how workflow-specific context can remain separated while selected approved facts are shared.

Handoff Design

Define what context a person or agent needs for a handoff without exposing unrelated or unauthorized material.

Access Rules

Specify read, write, approval, and retention permissions for each context domain before any shared-memory use.

Compaction Evaluation

Measure what summaries omit or distort before using compaction to control retrieval size and cost.

Source Relationships

Test whether linked sources improve retrieval while preserving provenance, contradictions, and uncertainty.

Workflow Relationship

Context must remain scoped and reviewable

This concept asks whether persistent context can improve reviewed handoffs across model and agent workflows without exposing unrelated data or reviving stale decisions.

The evaluation must separate useful retrieval from confident recall. Source provenance, permission boundaries, expiry rules, and human correction remain part of the system design.

Use Cases

Where persistent memory matters

Long-Running Work

A candidate use case for carrying approved decisions and source context across sessions without assuming recall is complete or current.

Cross-Session Context

A research question about when prior context improves a task and when stale context creates new errors.

Reviewed Handoffs

A candidate workflow for passing findings or prepared actions between specialized roles with explicit ownership.

Decision Records

An evaluation scope for recording the sources, retrieved context, proposed action, and approval attached to a consequential step.

FAQ

DomeAI, Common Questions

DomeAI is a research concept for persistent, reviewable context across AI agent sessions and workflows. This page does not represent a public cloud-memory product.
The proposed research adds retrieval ranking, source relationships, expiry, and task-specific permission checks around stored context. Each layer would need its own quality and failure evaluation.
Shared context is a candidate use case, not a validated capability. Any implementation would need explicit read, write, retention, and approval boundaries for each role.
No performance result is currently published. A credible evaluation would report retrieval quality and latency with the dataset size, index configuration, hardware, and cache state attached.
No standalone product is represented on this page. Contact us to discuss whether a scoped persistent-context evaluation fits a specific workflow.
Persistent Context

Need context that stays reviewable?

Let's define the sources, permissions, retention rules, retrieval measures, and approval boundaries for a scoped evaluation.

Fixed-price quote before any work starts
You own 100% of the code
30 days of free support
AI agents, custom software & automation
Fixed-price quotes, no hidden fees
30-day post-launch support