STATIC PROTOTYPE · NO LIVE PROVIDER CONNECTIONS

AI & memory, connected by provenance.

This workbench maps how personal context, augmented memories, source lineage, retrieval, consent and governance should fit together. It is a design and seed-data prototype, not a connected memory service.

Capture→Normalize→Connect→Rank→Resolve→Review / expire

Boundary: no provider memory is read or written; no account, live database, server-side persistence or consent enforcement is connected. Seed rows and ranking scores are illustrative.

Schema files

Context from capture through deployment.

Context dimensions

Identity, time, scope, meaning, relation, source, confidence, access, lifecycle.

Lifecycle states

Captured to expired or superseded.

Local

Data handling

Exports generated client-side.

Technical reference: Context model Schema suite Lifecycle engine Resolution sandbox Runtime surfaces Package & export

What counts as context?

Situation	Current task, activity, environment or event.
Actor & scope	Person, group, project, system, portfolio or resource the context concerns.
Intent	Goal, request, desired outcome, constraint or preference.
History	Prior events, decisions, commitments and outcomes relevant now.
Knowledge	Facts, concepts, definitions, evidence and known unknowns.
Temporal status	When it applied, when it was observed, and whether it remains current.
Provenance	Where it came from and whether it is directly observed, reported or inferred.

Resolution contract

A context resolution should answer:

What request or activity is being interpreted?
Which subject, project and time window apply?
Which context items are relevant, and why?
Which sources support them?
Are there conflicts, stale items or missing evidence?
What may be used, shown, retained or shared?
What context was supplied to the downstream action?

A retrieval result is a recommendation for use, not proof that the context is correct or permission to disclose it.

Schema suite

Thirteen portable CSV files form a starter package. They are a proposed schema baseline, not a claim that a production database or connected runtime already exists.

File	Purpose	Core fields
`01_CONTEXT_ITEMS.csv`	Canonical context item registry.	context_id, kind, content, subject_id, scope, status
`02_SOURCES.csv`	Source and provenance records.	source_id, source_type, locator, captured_at, trust_basis
`03_SUBJECTS.csv`	People, groups, projects, systems and resources.	subject_id, subject_type, label, namespace
`04_RELATIONSHIPS.csv`	Typed links between context entities.	source_id, target_id, relation, confidence
`05_TEMPORAL_VALIDITY.csv`	Observed time and valid time.	context_id, observed_at, valid_from, valid_to
`06_INTENTS_CONSTRAINTS.csv`	Goals, requests, preferences and limits.	item_id, subject_id, intent_type, priority, expiry
`07_EVIDENCE_LINKS.csv`	Evidence supporting or challenging claims.	context_id, evidence_id, relation, verification_status
`08_CONFIDENCE_ASSESSMENTS.csv`	Confidence and uncertainty rationale.	context_id, score, basis, assessed_at, assessor
`09_ACCESS_POLICIES.csv`	Use, visibility and retention constraints.	policy_id, scope, permitted_action, retention_rule
`10_LIFECYCLE_EVENTS.csv`	Append-only context change history.	event_id, context_id, event_type, timestamp, actor
`11_CONFLICTS.csv`	Contradictions and resolution records.	conflict_id, context_a, context_b, status, rationale
`12_RETRIEVAL_LOG.csv`	Why context was retrieved and used.	query_id, context_id, relevance, reason, outcome
`13_DEPLOYMENT_MANIFEST.csv`	Import order and validation checks.	artifact, path, load_order, target, validation

Relationship graph

RequestconcernsProjecthas_goalIntentconstrained_byConstraint

Context itemderived_fromSourcesupported_byEvidence

Context itemsupersedesEarlier itemmay_conflict_withOther item

Keep links typed and directional where meaning requires it. Do not silently merge entities merely because their labels look alike.

Controlled relationship vocabulary

`concerns`	Context applies to a subject.
`derived_from`	Provenance lineage.
`supported_by`	Evidence relation.
`contradicts`	Conflicting claims or values.
`supersedes`	Newer item replaces an earlier one for a stated scope.
`depends_on`	Interpretation requires another context item.
`relevant_to`	Item may inform a request or activity.
`visible_to`	Access scope, subject to policy validation.

Lifecycle engine

From	Event	To	Required check
captured	normalize	normalized	Stable ID and source
normalized	link	connected	Typed relationships validated
connected	assess	assessed	Time, confidence and access reviewed
assessed	approve use	available	Policy permits intended use
available	replace	superseded	Replacement link and scope
any active	expire / revoke	expired / revoked	Stop ordinary retrieval

Failure routes

Missing source	Mark provenance gap; restrict confidence.
Stale context	Check validity window or ask for confirmation.
Conflicting context	Preserve both claims; record conflict and resolution status.
Ambiguous subject	Do not merge; request disambiguation or retain separate IDs.
Access denied	Exclude from result and record policy outcome without exposing content.
Deletion / revocation	Block future retrieval and propagate deletion according to retention policy.

Context resolution sandbox

Enter a request and candidate context items to produce a transparent, illustrative relevance ranking. This demo uses simple keyword overlap only; it is not semantic AI retrieval and does not enforce real access controls.

Request / activitySubject or project scopeTime horizonCandidate context items (one per line)

Governance & evidence rules

These are requirements to implement and test, not backend controls enforced by this static prototype.

Preserve provenance and distinguish direct observation, reported statement and inference.
Use explicit observed-at and valid-time fields; do not assume the newest capture is always the newest truth.
Do not overwrite conflicting claims; retain conflict status and resolution rationale.
Apply access and retention rules before content enters a retrieval result.
Record why a context item was selected for an action and which version was supplied.
Support correction, expiry, revocation and deletion propagation.
Do not infer verified human identity from a matching name or shared identifier alone.

Quality signals

Provenance completeness	Has a source and capture details.
Temporal fitness	Valid for the time of the request.
Scope fit	Matches subject, project and activity.
Evidence strength	Supported by traceable evidence where needed.
Conflict status	Contradictions are visible, not silently resolved.
Permission fit	Intended use is permitted.
Retrieval explainability	Selection reason can be inspected.

Confidence values should be calibrated against an explicit evidence policy. A numeric score alone is not verification.

Capture surfaces

Where context enters

Conversation exports, forms, documents, project tools, event logs, user-provided preferences and system observations.

Resolution surfaces

Where context is used

Assistant prompts, search, workflow orchestration, planning, recommendations, agent tools and user-facing workbenches.

Evidence surfaces

Where context is checked

Source documents, audit events, timestamps, decision records, retrieval logs, receipts and correction histories.

User-facing control surface

OwnMyAI / OwnYourAI

OwnMyAI is the user-facing place to review browser-local context, audit user-supplied material, prepare a manual reset checklist and export a Context Passport.

The workspace saves only entries the user explicitly adds in that browser. It does not read or write provider-side memory.

Integration boundary

These are separate origins and products. This workbench cannot read OwnMyAI local storage, and opening a link does not share data between sites.

Provider read/write: not connected.
Cross-device sync and hosted memory: not connected.
Context Passport: selected browser-local entries only.
Provider reset: manual checklist only; no provider action is executed.

Proposed pipeline

Augmented memories

Source→Normalize→Link→Assess→Retrieve→Review

Augmentation should add typed relationships, temporal validity, task intent, constraints and source evidence without converting inference into fact.

Required safeguards

Keep the original source and the derived memory linked.
Separate explicit user statements from model inference.
Retain conflicts and stale versions instead of silently overwriting them.
Require a valid purpose and permission before retrieval or sharing.
Record which memory version informed an action and allow correction.

The pipeline is a proposed design contract. This static page does not run model-driven augmentation or enforce these safeguards.

Schema and lifecycle reference

Memory systems

The companion Memory Workbench explores the thirteen CSV schemas, lifecycle states and relationship graph. It is also a static prototype; it does not connect to a runtime or database.

This workbench and the companion schema explorer are separate static pages. Neither synchronises data with the other.

Runtime acceptance boundary

Capability	Current status
Authenticated memory CRUD	Not verified
Persistent multi-user store	Not connected
Cross-device synchronisation	Not connected
Source lineage and version history	Schema proposed
Provider memory import/write	Not connected

Permission boundary

Consent & permissions

The statuses below describe this prototype and its linked user-facing workspace; they are not a legal consent receipt or proof of provider-side enforcement.

Action	Status	What is actually verified
Save a context item	Browser-local	Explicit local-storage notice; no server sync or encryption by this app.
Audit supplied material	User-initiated	Supported user-supplied material only; not a provider API connection.
Read provider memory	Not connected	No provider memory is read by this workbench.
Write or delete provider memory	Not connected	Reset plan is an exported checklist; user must act in provider controls.
Revoke and propagate deletion	Not implemented	No live permission token, connector revocation or deletion propagation is wired here.

Connector inventory

Providers & sources

Provider names are shown as intended integration targets, not as active connections. No provider key or user credential is embedded in this static page.

Provider / source	Current path	Boundary
ChatGPT, Claude, Gemini	User-supplied audit material	No direct API read/write; completeness is not guaranteed.
OpenRouter and other model APIs	Not connected	No model call is made from this workbench.
Browser and local files	Schema/examples only here	No browser history or filesystem access from this static page.
Google Drive and project tools	Not connected	No remote import, search or sync.
Source lineage	Schema proposed	Example records are not evidence of a real source connection.

Package, seeds & deployment

context-workbench-package/ ├── 01_CONTEXT_ITEMS.csv ├── 02_SOURCES.csv ├── 03_SUBJECTS.csv ├── 04_RELATIONSHIPS.csv ├── 05_TEMPORAL_VALIDITY.csv ├── 06_INTENTS_CONSTRAINTS.csv ├── 07_EVIDENCE_LINKS.csv ├── 08_CONFIDENCE_ASSESSMENTS.csv ├── 09_ACCESS_POLICIES.csv ├── 10_LIFECYCLE_EVENTS.csv ├── 11_CONFLICTS.csv ├── 12_RETRIEVAL_LOG.csv ├── 13_DEPLOYMENT_MANIFEST.csv ├── context.graph.json └── README.md Suggested import order: 1. Subjects and sources 2. Context items and temporal validity 3. Relationships and intent/constraints 4. Evidence and confidence assessments 5. Access policies and lifecycle events 6. Conflicts and retrieval logs 7. Manifest validation and graph export Important: this static workbench does not connect to a live database, memory service, browser, or external account.